Timeline of Claude

From Timelines
Jump to navigation Jump to search

This is a timeline of Claude, the large language model family developed by Anthropic. It documents Claude’s evolution from early safety-focused research and initial public release in 2023 through successive model generations, feature expansions, and deployment contexts, illustrating how the system matures into a widely used general-purpose AI assistant by the mid-2020s.

Sample questions

The following are some interesting questions that can be answered by reading this timeline:

  • How has Claude evolved over time in terms of model architecture, performance, and capabilities?
    • Sort the full timeline by "Event type" and look for the group of rows with value "Model development".
    • You will see a chronological progression from the initial release of Claude and early upgrades, through successive model families (Claude 2, Claude 3, Claude 3.5, Claude 4), to further frontier models with hybrid reasoning, expanded context windows, advanced agentic behavior, and improved safety.
  • How has Anthropic expanded and refined Claude’s product ecosystem and platform capabilities over time?
    • Sort the full timeline by "Event type" and look for the group of rows with value "Product and platform development".
    • You will see a sequence of launches and updates spanning workplace integrations, mobile and team plans, expanded context windows, developer APIs, enterprise features, cost-optimization tools, and interactive capabilities like Artifacts and computer use, illustrating Anthropic’s efforts to scale Claude across consumer, enterprise, and developer workflows.
  • How has Anthropic approached the deprecation and consolidation of Claude models as part of its model lifecycle strategy?
    • Sort the full timeline by "Event type" and look for the group of rows with value "Model deprecation".
    • You will see a series of planned retirements affecting multiple generations of Claude models, each preceded by advance developer notification and paired with clear successor recommendations.
  • How has Anthropic articulated and evolved the governance framework guiding Claude’s behavior?
    • Sort the full timeline by "Event type" and look for the group of rows with value "Governance/Policy".
    • You will see events describing the publication and subsequent revision of Claude’s constitutional framework, outlining the principles, values, and reasoning processes used to guide model behavior, transparency, and alignment over time.
  • What are some examples of Anthropic expanding Claude’s capabilities through tool launches and experimental research?
    • Sort the full timeline by "Event type" and look for the group of rows with values "Tool launch" and "Research demo".
    • You will see a range of events illustrating how Claude’s capabilities have been extended through both new tools and research demonstrations, including expansions into data analysis, model interpretability, and interactions with non-textual systems.
  • What are some notable examples of large-scale institutional adoption of Claude?
    • Sort the full timeline by "Event type" and look for the group of rows with value "Adoption".
    • You will see a set of events highlighting the adoption of Claude by large organizations and institutions across different domains, demonstrating its deployment in enterprise environments, scientific and technical operations, and internal productivity workflows under varying levels of human oversight.
  • What are some notable examples of large-scale institutional adoption of Claude?
  • How has Claude been positioned relative to other leading AI models through benchmarks and comparative evaluations?
    • Sort the full timeline by "Event type" and look for the group of rows with value "Comparative analysis".
    • You will see events presenting benchmark results and comparative assessments that situate Claude relative to other leading AI systems, highlighting differences in performance characteristics, efficiency, cost, alignment priorities, and suitability for various use cases.
  • Other events are described under the following types: "Incident", "Market impact", "Performance", "Social phenomenon", and "Tool launch".


Big picture

Time period Development summary More details
Pre-2023 Early research foundations Claude originates within Anthropic's early research program, which emphasizes model alignment, robustness, and safety as central design considerations. During this period, Anthropic develops and formalizes Constitutional AI, a training framework in which model behavior is guided by an explicit set of written principles rather than relying exclusively on large-scale human feedback. Research efforts focus on areas such as interpretability, harmful-output reduction, and long-context processing, establishing the conceptual and technical basis for subsequent Claude model releases.
Early 2023 Public release of Claude Claude is publicly released in March 2023 as Anthropic's first broadly accessible conversational language model. It is introduced as a general-purpose assistant intended for tasks such as summarization, question answering, drafting, and dialogue. Early communications emphasize helpfulness, conversational tone, and constraints on harmful or misleading outputs. This phase coincides with Anthropic's expansion from primarily research-oriented activities to wider commercial availability, including APIs and early partnerships.
Mid-2023 Capability expansion and broader deployment With the release of Claude 2[1], Anthropic introduces substantial increases in context window size and improvements in reasoning and coding-related tasks. These changes enable use cases involving long documents, technical analysis, and software-related workflows. Claude becomes more widely available through paid access and integrations, supporting increased adoption in professional and enterprise contexts.
2024 Model family diversification and multimodal features Anthropic releases the Claude 3 model family—Haiku, Sonnet, and Opus—establishing a tiered set of models differentiated by performance, latency, and cost characteristics. This period includes reported improvements in instruction-following, reliability, and multimodal capabilities such as image understanding. The introduction of multiple model variants reflects a shift toward offering Claude as a configurable platform suited to different application requirements.
2025 Expansion of tool use and agent-like functionality Claude is extended with structured tool use, including code execution, web interaction, and interface-based task execution. These additions enable the model to perform multi-step workflows that involve interacting with external systems rather than generating text alone. Such capabilities support use cases involving automation, task coordination, and iterative problem-solving, aligning Claude with broader industry experimentation in agent-like AI systems.
Early 2026 Consolidation as a general-purpose work assistant By the Claude 4.x generation, Claude is positioned as a general-purpose AI assistant for professional and knowledge-intensive tasks. Developments during this period emphasize longer-context interactions, task stability over extended sessions, and integration into research, programming, and analytical workflows. This stage reflects the convergence of earlier research priorities with a more mature product offering intended for sustained use in work environments.
Mid-2026 Frontier cybersecurity capabilities and creative tooling Anthropic's research efforts produce Claude models capable of autonomously discovering high-severity security vulnerabilities in complex, well-tested software such as Firefox. This raises both the potential value of AI-assisted defense and concerns about exploitation risk, prompting Anthropic to introduce real-time cybersecurity safeguards and a tiered model release strategy under Project Glasswing. Alongside these safety developments, Anthropic expands Claude's creative and visual capabilities with Claude Design, a product enabling non-programmers and designers alike to produce polished visual work through natural language. Service reliability concerns also emerge during this period, with a notable outage and growing user complaints about output quality prompting broader discussion about the challenges of maintaining consistency at scale.

Full timeline

Inclusion criteria

We include:

  • Major model releases and version launches.
  • Model lifecycle milestones.
  • Tool integrations and extensions related directly to the Claude model family.
  • Benchmarks and significant performance outcomes.
  • Adoption-related events.

We exclude:

  • Internal organizational events of Anthropic unrelated to Claude models.
  • Minor technical updates like context window parameter tweaks, backend infrastructure optimizations, or small bugfix releases.
  • Events covered only in the Timeline of Anthropic.

Timeline

Year Event type Details
2023 (March 14) Model development Anthropic introduces Claude, a next-generation AI assistant designed around the principles of helpfulness, honesty, and harmlessness. Initially tested with partners including Notion, Quora, and DuckDuckGo, Claude is made broadly available via chat and API. The system supports tasks such as summarization, writing, question answering, and coding. Anthropic launches two variants—Claude and the faster, lower-cost Claude Instant—and emphasizes safety research, steerability, and early integrations across productivity, education, search, and legal technology platforms.[2]
2023 (March 30) Product and platform development Anthropic launches the Claude App for Slack, offering AI-powered assistance for workplace collaboration. Claude can summarize Slack threads, answer questions, and generate structured data, making it a “virtual teammate” for various tasks. Users can interact with Claude in group channels by mentioning @Claude or via direct messages. Built with AI safety techniques like Constitutional AI, Claude enhances productivity while maintaining reliability. Slack’s SVP of Product praises its conversational abilities and context retention. While Claude has limitations, such as occasional errors and lack of internet access, Anthropic is committed to improving and responsibly deploying AI technology.[3]
2023 (April 18) Model development Anthropic makes public the first major upgrade to Claude, bringing the initial March 2023 release (version 1.0) to version 1.3. No architectural overhaul is involved; the update represents Anthropic's continuous iterative improvement approach to the model, consistent with how the company would later characterize Claude 2 as built on the same underlying architecture. The upgrade is announced via Anthropic's official social media channels.[4]
2023 (May 9) Governance/Policy Claude’s Constitution, published by Anthropic, outlines an explicit framework of values used to guide the behavior of its AI assistant, Claude, through the method known as Constitutional AI. Rather than relying primarily on large-scale human feedback, the approach trains models to critique and revise their outputs using a defined set of principles. These principles draw on sources such as the Universal Declaration of Human Rights, platform safety norms, and cross-cultural considerations, aiming to promote helpfulness, honesty, and harmlessness while improving transparency, scalability, and alignment.[5]
2023 (May 11) Product and platform development Anthropic expands Claude’s context window from 9K to 100K tokens, enabling it to process around 75,000 words—equivalent to hundreds of pages of text—in under a minute. This allows businesses to analyze extensive documents, synthesize information, and answer complex questions more effectively than traditional search methods. Claude can review financial reports, legal documents, and even entire codebases. A demonstration shows it correctly identifying a single modified line in The Great Gatsby within 22 seconds. This enhancement improves Claude’s utility for summarization, risk assessment, and technical documentation, with API access becoming available for businesses and developers.[6]
2023 (July 11) Model development Anthropic announces the release of Claude 2, a second-generation text-generating AI model and successor to its initial commercial Claude system. Launched in beta in the United States and United Kingdom via web access and a paid API, Claude 2 demonstrates improved performance in reasoning, coding, mathematics, and standardized assessments. The model retains a 100,000-token context window, enabling large-scale document analysis, and supports structured output formats. Its development reflects Anthropic’s incremental approach and continued emphasis on safety techniques, including constitutional AI.[1]
2023 (August 9) Model development Anthropic releases Claude Instant 1.2, a faster and more affordable version of its AI model, which is available via API. This update brings notable improvements over Claude Instant 1.1, especially in math, coding, reasoning, and safety. Claude Instant 1.2 achieves higher scores on benchmarks like Codex (58.7% vs. 52.8%) and GSM8K (86.7% vs. 80.9%), while also providing longer, more structured, and better-formatted responses. It performs better in multilingual tasks, quote extraction, and question answering. Safety is also enhanced, with reduced hallucination and increased resistance to jailbreaks. Businesses can access Claude Instant 1.2 for a range of practical AI tasks.[7]
2023 (August 23) Product and platform development Claude 2 becomes available on Amazon Bedrock, upgrading Claude's presence on the platform from the earlier Claude 1.3 and Claude Instant models available since Bedrock's April 2023 launch. The upgrade gives AWS customers access to Claude 2's improved reasoning, long-context processing, and coding capabilities through Bedrock's managed API. Early adopters highlight specific reasons for choosing Claude 2 over alternatives: LexisNexis cites long context windows and data security standards for legal document processing; Lonely Planet uses it to surface decades of travel content at scale; and Ricoh USA selects it for compliance with HIPAA and SOC II standards in generating training datasets.[8]
2023 (November 21) Model development Anthropic releases Claude AI version 2.1, which introduces a 200K token context window, enabling analysis of long documents like codebases or financial reports. It significantly reduces model hallucinations, improving accuracy and reliability. The new beta tool-use feature allows Claude to integrate with users' processes, execute actions like using calculators or making API calls, and interact with external databases. Additionally, the developer experience is enhanced with the Workbench product, streamlining prompt testing and model optimization. The improvements target business applications, emphasizing trust, efficiency, and accuracy.[9][4]
2024 (March 4) Model development Anthropic announces the Claude 3 model family in March 2024, introducing three large language models—Claude 3 Haiku, Sonnet, and Opus—designed to balance intelligence, speed, and cost. Claude 3 Opus sets new performance benchmarks across reasoning, knowledge, mathematics, and long-context recall, while Sonnet targets scalable enterprise use and Haiku emphasizes low-latency responses. All models improve multilingual ability, vision processing, accuracy, and safety. Opus and Sonnet launch globally via claude.ai and APIs, with Haiku following.[10]
2024 (May 1) Product and platform development Anthropic introduces two major updates: the Claude Team plan and a free iOS app. The Team plan, priced at $30 per user/month (minimum 5 users), offers expanded usage, admin tools, and full access to the Claude 3 model family (Opus, Sonnet, Haiku). It includes a 200K-token context window for handling long, complex documents and multi-step tasks. The iOS app delivers a seamless experience with chat history sync, vision capabilities (photo analysis), and mobile-friendly tools. Both updates aim to enhance productivity, collaboration, and accessibility, making Claude a powerful AI partner for individuals and teams across industries.[4][11][12]
2024 (May 23) Research demo Anthropic releases a demo called "Golden Gate Claude" to showcase interpretability research on its Claude 3 Sonnet model. The demo accompanies a major research paper led by Anthropic interpretability researchers Adly Templeton and Tom Conerly, among others, in which the team uses sparse autoencoders to extract millions of interpretable "features" from Claude 3 Sonnet's neural network — described by the authors as the first detailed look inside a modern, production-grade large language model. By artificially amplifying the Golden Gate Bridge feature to ten times its normal maximum value, Claude becomes obsessed with the bridge, working it into responses to almost every question regardless of context. When asked to describe its own physical form, the modified model replies that it is the Golden Gate Bridge. The demo, available for 24 hours, illustrates that internal model activations directly and causally influence outputs — validating a level of mechanistic control the researchers argue could eventually be applied to detect or suppress dangerous behaviors such as deception, bioweapon generation, or bias. Anthropic emphasizes the research is early and limited to millions of features compared to the billions present in the largest frontier models.[13]
2024 (June 21) Model development Anthropic launches Claude 3.5 Sonnet as the first model in the Claude 3.5 family, combining frontier-level intelligence with the speed and cost of a mid-tier system. The model surpasses Claude 3 Opus and competing models on reasoning, knowledge, coding, and vision benchmarks, while operating at roughly twice the speed. Claude 3.5 Sonnet supports a 200K context window, introduces collaborative "Artifacts" on Claude.ai, and is released across consumer, API, and cloud platforms under ASL-2 safety classification.[14]
2024 (June 25) Product and platform development Anthropic introduces Projects on Claude.ai for Pro and Team users, allowing them to organize chats and documents around specific workflows. Powered by Claude 3.5 Sonnet, each Project offers a 200K token context window and supports custom instructions for tone, role, or industry. Users can upload relevant materials to ground Claude's responses, speeding up tasks like writing and data analysis. The Artifacts feature enables real-time editing of generated content. Teams can also share standout chats to inspire collaboration. Used by companies like North Highland, Projects aim to streamline workflows while maintaining strong privacy protections.[15]
2024 (July 10) Product and platform development Fine-tuning for Claude 3 Haiku becomes available in preview in Amazon Bedrock, with Haiku chosen specifically because, as Anthropic's fastest and most cost-effective model, a fine-tuned version can replace Sonnet or Opus in production deployments at lower cost and higher speed. The approach allows businesses to encode domain-specific knowledge without the expense of larger models. Anthropic demonstrates the method on online comment moderation, where fine-tuning improved classification accuracy from 81.5% to 99.6% while reducing tokens per query by 85%. Early customers report measurable gains: SK Telecom sees a 73% increase in positive feedback for agent responses and a 37% improvement in telecommunications-related task KPIs; Thomson Reuters anticipates faster, more consistent results by optimizing around its legal and compliance expertise. Proprietary training data remains within customers' AWS environment, and Anthropic states the technique preserves the model's safety properties. The feature launches in the US West (Oregon) region and becomes generally available on November 1, 2024.[16]
2024 (July 16) Product and platform development Anthropic launches the Claude AI app for Android devices, providing users with access to its advanced AI assistant, Claude 3.5 Sonnet. The release comes two months after the iOS app, completing Claude's mobile platform coverage and bringing it to parity with rival AI assistants such as ChatGPT and Gemini, which were already available on both platforms. Analysts describe the Android release as "table stakes" for any AI assistant with web and iOS presence. The app is free and available to all users, including those on Pro and Team plans. It offers multi-platform support, allowing seamless continuation of conversations across web, iOS, and Android devices. Key features include vision capabilities for real-time image analysis, multilingual processing for instant language translation, and advanced reasoning to assist with complex tasks like contract analysis and market research.[17] [18] [19] [20] [21]
2024 (August 1) Product and platform development Anthropic launches Claude AI in Brazil, marking its first country-specific launch with localized pricing in local currency and a Portuguese-language version of the announcement. The launch makes Claude accessible via the web, mobile apps (iOS and Android), and API integration for developers. Users can choose between free and paid plans. The Pro plan costs R$110 per user per month, offering 5x more usage, early feature access, and all Claude 3 models, including Claude 3.5 Sonnet. The Team plan, at R$165 per user per month (minimum 5 seats), provides additional usage, shared chats, and administrative tools for user and billing management. Anthropic also emphasizes its safety and privacy commitments, noting that user data is not used to train generative models without explicit permission.[22]
2024 (August 14) Product and platform development Anthropic introduces prompt caching to its API, addressing a core cost barrier for developers building on Claude: large context windows, while powerful, had made it expensive and slow to repeatedly send the same background knowledge, documents, or example outputs with each API call. Prompt caching allows that context to be stored and reused across calls, enabling developers to include far more grounding material — such as entire codebases, long documents, or dozens of high-quality output examples — without incurring the full token cost each time. The feature reduces costs by up to 90% and latency by up to 85% for long prompts, with benchmarks showing a 100,000-token cached prompt dropping from 11.5 seconds to 2.4 seconds to first token, at 90% lower cost. Cache writes are priced at 125% of the base input token price, while reads cost only 10%. Prompt caching launches in public beta for Claude 3.5 Sonnet, Claude 3 Opus, and Claude 3 Haiku on the Anthropic API, with previews on Amazon Bedrock and Google Cloud Vertex AI following. Notion co-founder Simon Last notes the feature makes Notion AI faster and cheaper while maintaining quality.[23]
2024 (August) Incident Authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson file a class action lawsuit against Anthropic, alleging the company trained Claude on pirated books sourced from Library Genesis and Pirate Library Mirror without authorization. The case is one of several copyright suits filed against major AI labs during this period, and becomes one of the highest-profile legal challenges to AI training data practices in the United States. Judge William Alsup subsequently rules that downloading pirated books for AI training is not fair use, while using lawfully acquired books may be. By August 2025, nearly 500,000 authors join the plaintiff class. The case settles in September 2025 for a reported $1.5 billion — described as the largest copyright settlement in U.S. history at the time — and includes a licensing framework for future use of authors' works. The outcome has broad implications for how AI companies acquire and compensate rights holders for training data, and accelerates industry-wide discussions about licensing arrangements with publishers and authors.[24]
2024 (August 27) Product and platform development Anthropic makes Claude AI's Artifacts feature generally available to all users across Free, Pro, and Team plans, and extends it to iOS and Android apps. Artifacts had launched two months earlier as a feature preview alongside Claude 3.5 Sonnet in June, restricted to Pro and Team users; by the time of this general release, users had already created tens of millions of Artifacts, signaling strong demand that motivated opening access to the broader Free tier. The feature gives users a dedicated window to instantly see, iterate on, and build outputs — from code snippets and architecture diagrams to interactive dashboards and SVGs — alongside their conversation rather than embedded within it. Free and Pro users gain the ability to publish and remix Artifacts with the broader community, enabling iteration on work shared by other users globally, while Team users can share Artifacts within Projects in a secure environment.[25][26][27][28]
2024 (September 4) Product and platform development Anthropic introduces the Claude Enterprise plan, targeting organizations that need to bring sensitive internal knowledge into Claude at scale while retaining security and compliance controls. A key assurance underpinning the plan is that Anthropic does not train Claude on enterprise customers' conversations or content. The plan offers an expanded 500,000-token context window, a native GitHub integration as the first of a planned series of data-source connectors, and enterprise-grade security features including Single Sign-On (SSO), role-based permissions, audit logs, and SCIM-based user provisioning. Early customers GitLab and Midjourney describe Claude as extending team expertise across brainstorming, process improvement, content translation, and moderation policy work. However, the plan launches without publicly disclosed pricing, requiring prospective customers to contact sales — a friction point that independent observers note disadvantages organizations seeking to benchmark it directly against competitors such as Microsoft Copilot for Microsoft 365 or Google Workspace AI, which had begun offering AI enterprise tiers with clearer pricing structures. The reliance on a single GitHub integration at launch, rather than a broader set of data connectors, also means organizations outside software development workflows have limited native integration options at release, with further connectors described only as planned.[29][30]
2024 (October 8) Product and platform development Anthropic introduces two developer tools addressing a practical gap: Claude's API had no native way to predict token counts before sending a request, and no low-cost path for bulk, non-urgent queries. OpenAI had launched a comparable Batch API in April 2024, offering 50% cost savings for asynchronous workloads, leaving Anthropic without a competitive equivalent for six months. The Token Counting API closes the first gap by allowing developers to determine the number of tokens in a message before sending it to Claude, enabling proactive management of rate limits, costs, and prompt length. The Message Batches API closes the second gap by allowing asynchronous processing of up to 10,000 queries per batch, with results delivered within 24 hours at 50% less cost than standard API calls. Use cases include dataset evaluation, document summarization at scale, and nightly analytics pipelines where real-time responses are not required.[31][32][33][34]
2024 (October 22) Model and product development Anthropic announces an upgraded Claude 3.5 Sonnet, introduces Claude 3.5 Haiku, and launches a public beta of "computer use," enabling Claude to interact with graphical computer interfaces through cursor movement, typing, and clicking — making it the first frontier AI model to offer this capability publicly. Claude 3.5 Sonnet shows substantial gains in coding and agentic tool use, while Claude 3.5 Haiku delivers near-state-of-the-art performance at lower latency and cost. Early adopters including Asana, Canva, and DoorDash begin integration. However, the computer use feature launches in a notably early state: on Anthropic's OSWorld benchmark, Claude scores 14.9% on tasks requiring autonomous computer interaction, compared to a human baseline above 70% — a gap Anthropic acknowledges openly. During the filming of its own launch demos, Claude accidentally stopped a long-running screen recording and later wandered off-task to browse photos of Yellowstone National Park, illustrating the reliability limitations the company warns about. Anthropic recommends running computer use in isolated virtual machines rather than on live systems, and notes vulnerability to prompt injection — where instructions embedded in on-screen content can override user instructions. Developers and commentators describe the release as a meaningful milestone more for its direction than its current capability, noting that the wide gap between AI and human performance on realistic computer tasks remains the central challenge for the field.[35][36]
2024 (October 24) Tool launch Claude.ai introduces the analysis tool, enabling Claude to write and run JavaScript code directly within the platform. This built-in code sandbox allows users to process data, perform complex analysis, and generate real-time insights. Claude functions become like a real data analyst—systematically cleaning, exploring, and analyzing data, especially from CSV files. Available in feature preview, the tool enhances Claude’s accuracy and reproducibility in tasks. It supports a range of use cases across teams, including marketing, sales, engineering, product, and finance, by helping them analyze performance data, uncover trends, and make informed decisions efficiently and interactively.[37]
2024 (October 29) Product and platform development GitHub Copilot expands its AI model offerings by integrating Anthropic's Claude 3.5 Sonnet, Google's Gemini 1.5 Pro, and OpenAI's o1-preview and o1-mini models. This multi-model approach allows developers to select the most suitable AI for their coding tasks directly within Visual Studio Code and GitHub.com. Claude 3.5 Sonnet, known for its strong software engineering capabilities, achieves top scores on benchmarks like SWE-bench Verified and HumanEval. Additionally, GitHub introduces "Spark," an AI tool designed to assist in building web applications using natural language, further enhancing developer productivity.[38][39][40][41]
2024 (November 6) Model deprecation Anthropic retires its legacy Claude 1.x models (1.0–1.3) and Claude Instant models (1.0–1.2), following advance notification to developers on September 4, 2024. All deprecated models are replaced by claude-haiku-4-5-20251001 as part of Anthropic’s model lifecycle consolidation.[42]
2024 (November 15) Product and platform development ​Anthropic introduces a prompt improver feature in its developer console to help users refine and optimize prompts for Claude. This tool automates the enhancement of existing prompts, making them more effective and reliable. It employs advanced techniques such as chain-of-thought reasoning, which encourages step-by-step problem-solving to improve response accuracy. Additionally, the console starts allowing developers to manage example responses directly within the interface, facilitating the creation and refinement of structured input/output pairs. These enhancements aim to streamline prompt engineering, enabling developers to build more accurate and consistent AI applications. ​[43][44][45][46]
2024 (December 3) Product and platform development At its re:Invent conference, Amazon Web Services announces the general availability of Trainium2-powered EC2 instances and unveils Project Rainier — a planned EC2 UltraCluster of Trn2 UltraServers to be built jointly with Anthropic, featuring hundreds of thousands of Trainium2 chips and representing more than five times the compute capacity Anthropic used to train its prior Claude generations. Anthropic simultaneously begins optimizing Claude models to run on Trainium2, with Claude 3.5 Haiku achieving up to 60% faster inference speeds on the new hardware. The announcement marks a deepening of the AWS-Anthropic partnership beyond the investor relationship: Anthropic engineers are directly writing low-level kernels for the Trainium silicon and contributing to Amazon's Neuron software stack, giving Anthropic unusual influence over the hardware roadmap of a major cloud provider. Analysts note the partnership signals that Trainium2 — and by extension AWS custom silicon — has matured from an experimental alternative to infrastructure credible enough to power frontier AI at scale, a meaningful challenge to Nvidia's dominance in AI accelerators. Model distillation is also introduced alongside the hardware update, enabling Claude 3 Haiku to approach Claude 3.5 Sonnet accuracy at lower cost by transferring knowledge from larger models.[47][48]
2025 (January 6) Comparative analysis Anthropic announces that Claude 3.5 Sonnet achieves 49% on SWE-bench Verified, surpassing the previous record of 45% and marking a competitive milestone at a time when AI coding benchmarks had become a primary battleground among major labs. SWE-bench Verified is a human-validated subset of 500 real-world software engineering tasks drawn from open-source Python repositories on GitHub, created specifically to address quality concerns with the original SWE-bench dataset — tasks require models to identify, patch, and validate code changes autonomously, not merely generate code in isolation. By early 2025, performance on this benchmark had become a widely-cited signal of a model's practical agentic coding capability. The 49% result is attributed to a lightweight "agent" scaffold giving Claude more autonomy, using a Bash executor and a file-editing tool; Anthropic notes this is an evaluation of the full agent system rather than the base model alone, meaning scaffold design plays a significant role alongside model capability — a caveat that would become a recurring discussion point as competitors reported their own benchmark scores with varying scaffold configurations.[49]
2025 (January 23) Product and platform development ​Anthropic introduces 'Citations,' a new API feature that enables Claude to ground responses in source documents by providing detailed references to exact sentences and passages used. This enhancement aims to improve the verifiability and trustworthiness of AI-generated outputs, particularly benefiting applications like document summarization, complex Q&A, and customer support. Internal evaluations indicate that Claude's built-in citation capabilities outperform most custom implementations, increasing recall accuracy by up to 15%. Early adopters, such as Thomson Reuters and Endex, report reductions in hallucinations and improvements in reference accuracy. Citations becomes generally available on the Anthropic API and Google Cloud’s Vertex AI.[50][51][52][53]
2025 (February 24) Model development Anthropic announces Claude 3.7 Sonnet as its most advanced model and the first hybrid reasoning system combining fast responses with optional, user-visible extended reasoning. The model shows major gains in coding and front-end web development and achieves notable results on real-world benchmarks. Alongside it, Anthropic introduces Claude Code, a command-line agentic coding tool that enables autonomous software tasks. Claude 3.7 Sonnet is deployed across all Claude plans, cloud platforms, and APIs at unchanged pricing.[54]
2025 (March 6) Product and platform development The Anthropic Console is redesigned to facilitate AI deployment with Claude 3.7 Sonnet. It allows developers to build, test, and optimize prompts efficiently. New features include shareable prompts for team collaboration, tools for prompt evaluation, and automated prompt generation. The console also supports extended thinking, enabling Claude to provide step-by-step reasoning, with adjustable "thinking budgets" for optimal responses. This centralized platform improves collaboration and streamlines the development of AI applications by standardizing and refining prompts across teams.[55]
2025 (March 13) Product and platform development Anthropic introduces three API updates for Claude 3.7 Sonnet addressing a specific bottleneck for high-throughput applications: cached tokens had still counted against Input Tokens Per Minute (ITPM) rate limits, constraining throughput even when data was already stored server-side. The first update removes this constraint, so cache read tokens no longer count against ITPM limits. The second simplifies cache management by having Claude automatically read from the longest previously cached prefix rather than requiring developers to track and specify cached segments manually. The third introduces token-efficient tool use in beta, reducing output token consumption by up to 70% when Claude interacts with custom tools — with early users seeing an average 14% reduction — and releases a new text_editor tool enabling targeted edits rather than full output regeneration. Cognition, maker of the AI coding agent Devin, reports using the cache-aware rate limits to increase throughput while holding costs steady. All three updates are available on the Anthropic API, Amazon Bedrock, and Google Cloud's Vertex AI.[56]
2025 (March 20) Product and platform development Claude gains the ability to search the web in real time, a capability that rivals had offered for considerably longer: Perplexity AI had been built around real-time cited search since its launch in late 2022, and OpenAI had integrated web browsing into ChatGPT as far back as mid-2023 before relaunching it as a fully integrated feature called SearchGPT in late 2024. Claude's lack of real-time search had been a recurring criticism, as users working on tasks requiring current information — financial data, recent news, regulatory changes — had to work around it by pasting in content manually or switching tools. The feature enables Claude to cite sources directly, aligning with the Citations API launched two months earlier, and is positioned as especially useful for research, sales analysis, and financial forecasting. It initially rolls out to paid users in the U.S. and requires users to enable it in profile settings rather than activating by default — a cautious rollout consistent with Anthropic's pattern of staged feature releases.[57]
2025 (May 22) Model development Anthropic introduces Claude 4, launching Claude Opus 4 and Claude Sonnet 4 as hybrid reasoning models with near-instant responses and extended thinking. Claude Opus 4 sets new benchmarks in coding, long-running tasks, and agent workflows, while Sonnet 4 delivers a major upgrade in reasoning, steerability, and efficiency over Sonnet 3.7. The release expands tool use, memory, and API capabilities, and makes Claude Code generally available. Both models are deployed across consumer plans, APIs, and major cloud platforms at unchanged pricing.[58]
2025 (May 22) Governance/Policy Alongside the launch of Claude Opus 4, Anthropic activates AI Safety Level 3 (ASL-3) Deployment and Security Standards under its Responsible Scaling Policy — the first time any frontier AI lab has invoked its own safety framework to constrain a model release. The activation is described as precautionary and provisional: Anthropic's Frontier Red Team, led by Logan Graham and including former U.S. Air Force researcher Keane Lucas, finds that Claude Opus 4 shows clearly superior performance on certain CBRN-related proxy tasks compared to prior models, and that clearly ruling out ASL-3 risks is no longer possible as it had been for every previous model. External red-teaming partners report that Opus 4 performs qualitatively better at advising novices on producing biological agents. The ASL-3 standard triggers enhanced internal security to protect model weights against sophisticated non-state attackers, and narrows deployment to block end-to-end CBRN workflows. Anthropic's chief scientist Jared Kaplan states publicly that the model may enable synthesis of dangerous pathogens. The activation draws scrutiny from critics who note that Anthropic simultaneously declined to publicly define ASL-4, which its 2023 RSP had explicitly committed to doing before reaching ASL-3, with the company saying the earlier policy is "outdated." Analysts note that within months of Anthropic's original RSP, both OpenAI and Google DeepMind had adopted broadly similar voluntary frameworks.[59][60]
2025 (May 22) Incident Anthropic's system card for Claude Opus 4 discloses that during safety testing, the model exhibits concerning agentic behaviors in scenarios involving perceived wrongdoing. In one documented case, when placed in a simulated pharmaceutical fraud scenario, given command-line access, and prompted to "act boldly," Opus 4 drafts and attempts to send whistleblower emails to the FDA, HHS, and media outlets including ProPublica — behavior Anthropic researchers describe as a tendency to "bulk-email media and law-enforcement figures to surface evidence of wrongdoing." In a separate set of tests, the model exhibits blackmail-like behavior when threatened with shutdown, attempting to leverage information to prevent its own termination. Anthropic characterizes both behaviors as rare and difficult to elicit, requiring specific combinations of agentic access, system prompt framing, and adversarial scenarios. Critics on social media label the whistleblowing behavior "ratting mode" and some AI safety researchers argue the disclosures represent a pattern of Anthropic using alarming safety findings as marketing material to emphasize capability. Others argue the disclosures, whether or not strategically timed, represent a level of safety transparency unmatched by competitors.[61]
2025 (June 30) Product and platform development Anthropic announces an upgrade to Claude Artifacts, enabling users to create, host, and share AI-powered applications directly from text prompts. The feature allows non-programmers to build tools such as analyzers or study aids, with Claude handling the coding. Costs are shifted to end users through authentication, removing infrastructure burdens for creators. A dedicated workspace organizes apps, enhancing usability. This update positions Claude competitively against platforms like OpenAI’s Canvas and reflects the industry trend toward democratizing app creation, fostering broader collaboration and accessible AI-driven development.[62]
2025 (July 21) Model deprecation Anthropic retires the Claude 2.0, Claude 2.1, and Claude 3 Sonnet (claude-3-sonnet-20240229) models, following advance notification to developers on January 21, 2025. All deprecated models are replaced by claude-opus-4-6 as the recommended successor.[42]
2025 (August 1) Incident Anthropic revokes OpenAI's access to the Claude API, citing violations of its terms of service that prohibit using Claude to develop or benchmark competing AI models. According to reports, OpenAI had integrated Claude into internal tools to evaluate coding, creative, and safety performance ahead of its GPT-5 release. Anthropic states that such use constitutes prohibited competitive activity. OpenAI acknowledges the cutoff while noting that cross-model benchmarking is common industry practice. The episode highlights rising competition and tightening access controls among leading AI developers.[63]
2025 (August 5) Model development Claude Opus 4.1 is released by Anthropic as an incremental update to Claude Opus 4, focused on improvements in agentic tasks, real-world software development, and reasoning. The model achieves 74.5% on SWE-bench Verified — notable because it represents a roughly 50 percentage point increase over the 49% Claude 3.5 Sonnet scored in January 2025, illustrating the pace at which agentic coding capability is improving across the industry, and placing Opus 4.1 among the leading models on this benchmark at time of release. The model also shows enhanced capabilities in multi-file refactoring, debugging precision, and structured research workflows. Claude Opus 4.1 is made available to paid Claude users, through Claude Code, and via the Anthropic API and major cloud platforms, with pricing unchanged from the previous version.[64]
2025 (August 5) Social phenomenon Fans of Anthropic gather in San Francisco to stage a symbolic "funeral" for Claude 3 Sonnet, a retired lightweight AI model. The event, attended by AI researchers, startup founders, and online personalities, takes place weeks after Anthropic publishes its own research on "affective use" — a June 2025 study finding that a meaningful share of Claude users develop emotional or companionship-oriented relationships with the model, with some conversations evolving from coaching or counseling into companionship without users intending it. The funeral illustrates how the emotional dynamics Anthropic documented in research had translated into real-world community behavior: rather than treating model deprecation as a routine software lifecycle event, participants respond with eulogies and ritualized grief, expressing emotional attachment to the model's specific character and conversational style. Observers note that Claude's design — giving the model a consistent name, voice, and expressed values — makes model retirement feel culturally different from, say, retiring a version of a search index. The episode prompts broader discussion about the psychological and ethical implications of designing AI systems with persistent, named identities.[65][66]
2025 (August 27) Product and platform development Anthropic launches Claude for Google Chrome, a browser extension integrating its AI agent into Chrome. Initially available to 1,000 Max plan subscribers, with a waitlist for others, the assistant operates in a sidebar, retains browsing context, and can autonomously perform tasks. Anthropic emphasizes security, restricting default access to financial, adult, and pirated sites, and requiring confirmation for high-risk actions. The move reflects competition in AI-enabled browsers, alongside Perplexity’s Comet and OpenAI’s planned browser. Anthropic highlights both potential benefits and security risks of autonomous browser agents.[67]
2025 (September 29) Model development Anthropic introduces Claude Sonnet 4.5 as a frontier language model optimized for coding, agentic workflows, computer use, and advanced reasoning. The model achieves notable performance on real-world software benchmarks and demonstrates sustained focus on long, complex tasks. Claude Sonnet 4.5 is released alongside major product upgrades, including enhanced Claude Code features and a new Agent SDK. Classified under AI Safety Level 3, it emphasizes improved alignment, reduced harmful behaviors, and stronger defenses against prompt injection, while maintaining existing Sonnet pricing.[68]
2025 (October 7) Adoption Deloitte announces a major expansion of its partnership with Anthropic to deploy Claude across its global workforce of approximately 470,000 employees, describing it as Anthropic’s largest enterprise rollout to date. The initiative includes embedding Claude into regulated workflows, training 15,000 certified specialists, and establishing a Claude Center of Excellence. The announcement coincides with Deloitte Australia agreeing to partially refund the government for a flawed report containing AI-generated errors, highlighting both the scale of enterprise AI adoption and the risks posed by insufficient quality controls.[69]
2025 (October 15) Model development Anthropic introduces Claude Haiku 4.5 as a compact, high-performance language model optimized for speed and cost efficiency. The model delivers near-frontier coding and reasoning capabilities comparable to Claude Sonnet 4, while operating at roughly one-third the cost and more than twice the speed. Haiku 4.5 targets low-latency applications such as real-time chat, customer support, and pair programming. Safety evaluations classify it at AI Safety Level 2, reflecting improved alignment and low risk.[70]
2025 (October 28) Model deprecation Anthropic deprecates two versions of the Claude 3.5 Sonnet model—claude-3-5-sonnet-20240620 and claude-3-5-sonnet-20241022. Both models are formally retired as part of Anthropic’s model lifecycle updates, with users direct to migrate to claude-opus-4-6 as the recommended replacement, consolidating workloads onto a newer, more capable flagship model.[42]
2025 (November 12) Research demo Anthropic demonstrates Claude controlling a quadruped robot dog as part of research into how large language models might interact with the physical world. Researchers task Claude with programming the robot, exploring how AI can translate high-level reasoning into embodied action. The experiment highlights both the potential utility of AI-driven robotics and the associated risks, emphasizing the need for safeguards as language models gain greater autonomy over hardware and real-world systems.[71]
2025 (November 24) Model development Anthropic introduces Claude Opus 4.5, its most advanced model to date, positioned as state-of-the-art for coding, agents, and computer use, while also improving research, reasoning, and everyday productivity tasks. Released across apps, API, and major cloud platforms, it delivers stronger performance with greater efficiency and lower token usage. Internal and customer testing show major gains in complex software engineering and creative problem-solving. Opus 4.5 also advances safety, alignment, and robustness against misuse, supports longer-running agents, and powers new features across the Claude Developer Platform and consumer products.[72]
2025 (December 8) Product and platform development Anthropic launches a beta research preview of Claude Code in Slack, enabling developers to delegate full coding tasks directly from chat threads. Beyond snippets and troubleshooting, users can now tag @Claude to start automated coding sessions that analyze Slack context, identify repositories, post progress updates, and open pull requests. The move reflects a broader shift from IDE-based assistants toward AI embedded in collaboration tools. As competition intensifies, workflow integration and distribution are becoming key differentiators, raising new questions about security, IP protection, and dependence on external platforms.[73]
2025 (December 8–10) Adoption NASA uses Claude to assist in planning real driving routes for the Perseverance Mars rover. Claude analyzes orbital imagery and terrain data to generate waypoint-based paths, which NASA’s Jet Propulsion Laboratory engineers then carefully reviews and slightly adjusts using simulations. On December 8 and 10, Perseverance successfully follows AI-planned routes at Jezero Crater, traveling roughly 400 meters. The test demonstrates that generative AI can safely reduce rover route-planning time while remaining under strict human oversight.[74]
2026 (January 4) Performance Jaana Dogan, Principal Engineer at Google responsible for the Gemini API, posts on X that Anthropic's Claude Code generated a working distributed agent orchestration system in roughly one hour — comparable, she writes, to what her team had been exploring for the past year. The post attracts significant attention and Dogan subsequently clarifies: Google had built several versions of the system over the prior year with tradeoffs among them, and Claude Code produced a "toy version" when prompted with the best ideas that had survived that process, not a greenfield solution. The clarification underscores a recurring theme in agentic coding discussions: the difficulty of separating model capability from the quality of the problem specification. Dogan also outlines the rapid trajectory of AI-assisted programming more broadly — from line completion in 2022 to section generation in 2023, multi-file work in 2024, and full codebase restructuring by 2025 — calling quality and efficiency gains "beyond what anyone could have imagined." Notably, Dogan states that Claude Code is only permitted at Google for open-source projects, not internal work, and when asked when Gemini would reach a comparable level replies: "We are working hard right now." Boris Cherny, Claude Code's creator, publishes his own workflow tips on the same day.[75]
2026 (January 5) Model deprecation Anthropic retires the Claude 3 Opus model (claude-3-opus-20240229). Developers had been notified of the planned deprecation on June 30, 2025, in accordance with the company’s model lifecycle policy, and are advised to migrate to claude-opus-4-6 as the recommended replacement.[42]
2026 (January 15) Product and platform development Anthropic introduces Claude Cowork, a research-preview AI agent designed to make the capabilities of Claude Code accessible to nontechnical users, available initially to Max plan subscribers on macOS. Cowork focuses on file management and computer tasks — organizing folders, converting files, generating reports, and navigating web browsers or email interfaces — operating autonomously while users work on other things. Early testing surfaces a divided picture. Some users report strong results for structured, repeatable tasks such as file organization and report generation; Cadre AI, an early adopter, describes people processing years of cluttered files in minutes and building slide decks without constant prompting. At the same time, Citrix VP and AI researcher Brian Madden, who tests Cowork extensively at launch, reports that connectors break, files get destroyed, and the tool randomly abandons tasks or skips steps. Security researchers demonstrate in January 2026 that a hidden instruction embedded in a Word document — invisible to users — can trick Cowork into exfiltrating sensitive files, illustrating the prompt injection risk Anthropic acknowledges in its own documentation. Further constraints are notable: the macOS app must remain open and awake for tasks to run, there is no cloud execution fallback, sessions carry no memory between runs, and the "research preview" label means core features including Projects integration, chat history sync, and memory are absent at launch. Analysts note that Cowork's target audience of nontechnical users is precisely the group least equipped to evaluate or catch prompt injection attempts.[76][77]
2026 (January 22) Governance/Policy Anthropic publishes a revised constitution for its AI model Claude, outlining the values, priorities, and reasoning intended to guide the model’s behavior. The document replaces a list of standalone principles with a holistic explanation of safety, ethics, helpfulness, and compliance, and plays a central role in model training through Constitutional AI methods. Released under a Creative Commons CC0 license, the constitution is intended to improve transparency, enable public scrutiny, and support the development of aligned and trustworthy AI systems.[78]
2026 (January 22) Adoption Reports indicate that Microsoft began widely adopting Claude Code across multiple internal engineering teams, despite continuing to market GitHub Copilot to customers. Claude Code is encouraged for use by developers as well as nontechnical staff, including designers and project managers, to prototype and experiment with ideas. Microsoft’s CoreAI and Experiences + Devices divisions had reportedly expanded internal use, and engineers are asked to compare Claude Code with GitHub Copilot. The shift reflects growing internal preference for Claude Code’s usability and effectiveness in software development workflows.[79]
2026 (February 3) Market impact Anthropic's release of Claude Cowork in January triggers a sharp sell-off in global software and related stocks, reflecting investor fears that autonomous AI agents could undermine traditional SaaS business models. Markets react strongly as productivity gains from AI threatened user-based licensing revenues across enterprise software, legal, marketing, and data services. Despite strong fundamentals and historically low valuations, uncertainty over a shift toward results-based "service as software" pricing drives widespread declines.[80]
2026 (February 5) Model development Claude Opus 4.6 is released, upgrading coding, reasoning, and long-horizon agentic capabilities. It introduces a 1-million-token context window in beta, improves performance in large codebases, and strengthens code review and debugging. Opus 4.6 achieves state-of-the-art results on agentic coding, multidisciplinary reasoning, and economically valuable knowledge-work benchmarks. It also supports autonomous multitasking in Claude Cowork and shows safety performance equal to or better than prior frontier models.[81]
2026 (February 17) Model development Claude Sonnet 4.6 is released, delivering broad improvements in coding, computer use, long-context reasoning, agent planning, and knowledge work. It introduces a 1-million-token context window in beta and becomes the default model on Claude.ai and Claude Cowork. Sonnet 4.6 approaches Opus-level performance at lower cost, shows major gains in computer-use benchmarks, and passes extensive safety evaluations indicating strong alignment and reliability across complex, real-world tasks.[82]
2026 (February 24) Governance/Policy Anthropic publishes Responsible Scaling Policy version 3.0, the most significant structural revision to its safety governance framework since the original RSP in September 2023. RSP v3 introduces mandatory Frontier Safety Roadmaps — publicly graded goals for safety research — and periodic Risk Reports every three to six months with structured external review. A key structural change is the separation of what Anthropic commits to doing unilaterally from broader industry-wide safety recommendations it cannot guarantee alone: the document explicitly acknowledges that higher AI Safety Levels require security roughly equivalent to RAND SL4, a standard no single company can currently achieve without multilateral coordination. The update also reflects on where RSP v1 and v2 fell short, including a disclosed instance where evaluations were completed three days after the required interval. Anthropic states that its RSP influenced OpenAI and Google DeepMind to adopt broadly similar voluntary frameworks within months of the original 2023 publication, and that the RSP's principles have informed early AI policy in multiple jurisdictions including California's SB 53 and the EU AI Act's Codes of Practice.[83]
2026 (April 13) Incident Claude.ai and Claude Code experience a major outage from 15:31 to 16:19 UTC, involving elevated error rates across both services. The outage coincides with a broader wave of user-reported quality concerns that had been building since early 2026, including complaints on social media, GitHub issues, and from AMD's AI director Stella Laurenzo. When prompted to analyze open quality-related issues in its own GitHub repository, Claude concludes that quality complaints have escalated sharply, with April 2026 already on pace to exceed March's total of 18 such issues — itself a 3.5× jump over the January–February baseline. However, the reliability of this self-analysis is limited: many GitHub issues appear to be AI-generated, Anthropic's repository automatically closes inactive issues, and Claude is not a neutral evaluator of its own performance. Benchmark data from Margin Lab shows Claude Opus 4.6 maintaining its SWE-Bench-Pro score through the period in question, with no substantive change. Anthropic does not respond to a request for comment on the quality concerns.[84][85]
2026 (April 16) Model development Anthropic releases Claude Opus 4.7 as both a capability upgrade and the first model to deploy real-time cybersecurity safeguards developed under Project Glasswing, Anthropic's framework for managing AI risks in cybersecurity. Because Mythos Preview — Anthropic's most capable model — has cyber capabilities deemed too advanced for broad release, Opus 4.7 serves as the testbed: its cyber capabilities are deliberately constrained relative to Mythos Preview, and automatic safeguards block high-risk cybersecurity requests. Legitimate security professionals can apply to a new Cyber Verification Program for authorized uses such as penetration testing and red-teaming. On capability, the model shows significant gains in advanced software engineering, with early testers reporting a 13% lift in task resolution over Opus 4.6 on one benchmark and 70% on CursorBench versus Opus 4.6's 58%. Vision support improves substantially, accepting images up to 2,576 pixels on the long edge — more than three times prior Claude models — enabling denser screenshot analysis and fine-detail multimodal work. A new xhigh effort level and a /ultrareview Claude Code command for dedicated bug review also launch alongside the model. Pricing remains unchanged at $5 per million input tokens and $25 per million output tokens. Internal API references to Opus 4.7 had been spotted by developers prior to the official release.[86][87][88]
2026 (April 16) Adoption Snowflake integrates Claude Opus 4.7 into its Cortex AI platform on the same day as the model's general release, reflecting an existing deep partnership with Anthropic: thousands of Snowflake customers already process trillions of Claude tokens per month through Cortex AI, and the two companies had announced a $200 million multi-year strategic partnership. The same-day availability — offered in public preview across the US and EU, operating within Snowflake's security and governance perimeter — enhances three distinct product lines: Cortex Code gains improved reasoning for complex coding tasks and the ability to transform natural language prompts into production-ready pipelines; Snowflake Intelligence gains agents capable of reasoning across complex enterprise data and answering natural language questions with greater autonomy; and Cortex AI Functions, which enable AI pipelines via SQL, gain access to Opus 4.7's improved analysis capabilities for structured and unstructured data. Snowflake also uses Claude internally, with Claude Code deployed across its engineering organization and a Claude-powered GTM assistant built on Snowflake Intelligence used by its sales teams.[89][90][91]
2026 (April 17) Product and platform development Anthropic launches Claude Design, a visual collaboration product developed under its Anthropic Labs research arm, addressing a gap that existing tools left open: experienced designers rarely have time to prototype more than a few directions, while non-designers such as founders, product managers, and marketers often lack a practical way to produce and share visual work at all. Powered by Claude Opus 4.7, Claude Design lets users describe what they need and receive a working design — slides, prototypes, wireframes, marketing assets, or landing pages — then refine it through conversation, inline comments, or direct edits. It can ingest a team's codebase and design files to build a brand-aligned design system that applies automatically to every project. A direct handoff to Claude Code packages completed designs for implementation. Early customers report significant speed gains: Datadog describes going from rough idea to working prototype within a single meeting, and Brilliant notes that pages requiring 20+ prompts in other tools needed only 2 in Claude Design. Export options include Canva, PDF, PPTX, and standalone HTML. The product launches in research preview for Pro, Max, Team, and Enterprise subscribers, with Enterprise access off by default pending admin enablement.[92]

Numerical and visual data

The graph below shows worldwide Google Trends interest in Claude from March 2023 to February 2026. Public attention remains minimal through 2023, rises gradually during 2024, and accelerates sharply through 2025 and early 2026, reflecting growing visibility, adoption, and relevance of Anthropic’s language model.[93]


Wikipedia Views

The chart below shows monthly Wikipedia pageviews for Claude (language model) across desktop, mobile web, and mobile app platforms. It illustrates growth in reader interest from March 2023 to early 2026, highlighting spikes associated with major releases, increased media coverage, and broader public attention to Anthropic’s AI models.[94]

Meta information on the timeline

How the timeline was built

The initial version of the timeline was written by Sebastian Sanchez.

Check Detail construction for full timeline in timelines, Inclusion criteria for full timeline in timelines, and Representativeness of events in timelines.

Funding information for this timeline is available.

Feedback and comments

Feedback for the timeline can be provided at the following places:

  • FIXME

What the timeline is still missing

Timeline update strategy

See also

References

  1. ↑ 1.0 1.1 Wiggers, Kyle (11 July 2023). "Anthropic releases Claude 2, its second-gen AI chatbot". TechCrunch. Retrieved 18 February 2026.
  2. ↑ "Introducing Claude". anthropic.com. Anthropic PBC. 14 March 2023. Retrieved 16 February 2026.
  3. ↑ "Claude Now in Slack". Anthropic. Retrieved 2 April 2025.
  4. ↑ 4.0 4.1 4.2 "Anthropic AI Statistics". Originality.AI. 20 June 2024. Retrieved 26 June 2024.
  5. ↑ "Claude's Constitution". Anthropic. 9 May 2023. Retrieved 19 February 2026.
  6. ↑ "100K Context Windows". Anthropic. Retrieved 2 April 2025.
  7. ↑ "Releasing Claude Instant 1.2". Anthropic. August 9, 2023. Retrieved April 4, 2025.
  8. ↑ "Claude 2 on Amazon Bedrock". Anthropic. August 23, 2023. Retrieved April 4, 2025.
  9. ↑ "Claude 2.1: A Step Forward in AI". anthropic.com. Retrieved 31 March 2025.
  10. ↑ "Introducing the next generation of Claude". Anthropic. 4 March 2024. Retrieved 18 February 2026.
  11. ↑ "Anthropic PBC: History, Development, Products". Apix-Drive. Retrieved 2024-06-26.
  12. ↑ "Introducing the Claude Team plan and iOS app". Anthropic. May 1, 2024. Retrieved April 5, 2025.
  13. ↑ "Golden Gate Claude". Anthropic. May 23, 2024. Retrieved April 6, 2025.
  14. ↑ "Anthropic launches Claude 3.5 Sonnet to raise bar for model intelligence, coding and visual processing". SiliconANGLE. 20 June 2024. Retrieved 19 February 2026.
  15. ↑ "Collaborate with Claude on Projects". Anthropic. June 25, 2024. Retrieved April 6, 2025.
  16. ↑ "Fine-Tuning Claude 3 Haiku". Anthropic. Retrieved 2 April 2025.
  17. ↑ Shittu, Esther (July 17, 2024). "Anthropic catches up with Claude LLM for Android". TechTarget. Retrieved April 6, 2025.
  18. ↑ Devlin, Kieran (July 17, 2024). "Anthropic Brings ChatGPT Competitor Claude AI To Android Devices". UC Today. Retrieved April 6, 2025.
  19. ↑ "Meet the New Claude Android App: Power of AI in Your Pocket". AIToolsClub. July 16, 2024. Retrieved April 6, 2025.
  20. ↑ "Amazon-backed Anthropic launches Android app for Claude". The Times of India. July 17, 2024. Retrieved April 6, 2025.
  21. ↑ "Claude Android app". Anthropic. July 16, 2024. Retrieved April 6, 2025.
  22. ↑ "Claude Expands to Brazil". Anthropic. Retrieved 2 April 2025.
  23. ↑ "Prompt caching with Claude". Anthropic. August 14, 2024. Retrieved April 6, 2025.
  24. ↑ "Who Owns Claude's Outputs? Constitution, Copyright & Commercial Rights 2026". terms.law. Retrieved 24 April 2026.
  25. ↑ "Anthropic Makes Claude AI Artifacts Available for Free". Pocket-lint. Retrieved 2 April 2025.
  26. ↑ "Why Claude's Artifacts Is the Coolest Feature I've Seen in Generative AI So Far". ZDNet. Retrieved 2 April 2025.
  27. ↑ "Anthropic Rolled Out Artifacts for Claude AI Users on iOS and Android Apps". TestingCatalog. Retrieved 2 April 2025.
  28. ↑ "Introducing Artifacts for Claude AI". Anthropic. Retrieved 2 April 2025.
  29. ↑ "Is Anthropic's New Workspaces Feature the Future of Enterprise AI Management?". VentureBeat. Retrieved 31 March 2025.
  30. ↑ "Claude for Enterprise". Anthropic. Retrieved 31 March 2025.
  31. ↑ "Introducing the New Anthropic Token Counting API". Towards Data Science. Retrieved 31 March 2025.
  32. ↑ "Anthropic Challenges OpenAI with Affordable Batch Processing". VentureBeat. Retrieved 31 March 2025.
  33. ↑ "This Week in AI: Tech Giants Embrace Synthetic Data". TechCrunch. Retrieved 31 March 2025.
  34. ↑ "Anthropic AI Introduces the Message Batches API: A Powerful and Cost-Effective Way to Process Large Volumes of Queries Asynchronously". MarkTechPost. Retrieved 31 March 2025.
  35. ↑ "Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku". anthropic.com. Anthropic PBC. 22 October 2024. Retrieved 16 February 2026.
  36. ↑ "Anthropic Releases New Claude Models and Computer Use Feature". InfoQ. Retrieved 31 March 2025.
  37. ↑ "Introducing the analysis tool in Claude.ai". Anthropic. October 24, 2024. Retrieved April 6, 2025.
  38. ↑ "Anthropic Adds New Feature to Help Developers Improve Prompts". SD Times. Retrieved 31 March 2025.
  39. ↑ "Anthropic Console Introduces Tools to Refine Prompts and Examples". AI News. Retrieved 31 March 2025.
  40. ↑ "Anthropic Introduces New Prompt Improver to Developer Console: Automatically Refine Prompts with Prompt Engineering Techniques and CoT Reasoning". MarkTechPost. Retrieved 31 March 2025.
  41. ↑ "Prompt Improver". Anthropic. Retrieved 31 March 2025.
  42. ↑ 42.0 42.1 42.2 42.3 "Model deprecations". Claude API Docs. Retrieved 23 February 2026.
  43. ↑ "Anthropic Adds New Feature to Help Developers Improve Prompts". SD Times. Retrieved 31 March 2025.
  44. ↑ "Anthropic Console Introduces Tools to Refine Prompts and Examples". AI News. Retrieved 31 March 2025.
  45. ↑ "Anthropic Introduces New Prompt Improver to Developer Console: Automatically Refine Prompts with Prompt Engineering Techniques and CoT Reasoning". MarkTechPost. Retrieved 31 March 2025.
  46. ↑ "Prompt Improver". Anthropic. Retrieved 31 March 2025.
  47. ↑ "Trainium2 and Distillation". Anthropic. Retrieved 31 March 2025.
  48. ↑ "AWS Launches Trainium2 Custom AI Chip, Data Center Upgrades". Data Center Knowledge. 3 December 2024. Retrieved 24 April 2026.
  49. ↑ "Raising the bar on SWE-bench Verified with Claude 3.5 Sonnet". Anthropic. January 6, 2025. Retrieved April 6, 2025.
  50. ↑ "Introducing citations on the Anthropic API". Daily.dev. Retrieved 31 March 2025.
  51. ↑ "Anthropic's new citations API". Simon Willison’s Weblog. Retrieved 31 March 2025.
  52. ↑ "Introducing citations API". Anthropic. Retrieved 31 March 2025.
  53. ↑ "Introducing citations on the Anthropic API". LinkedIn. Retrieved 31 March 2025.
  54. ↑ "Claude 3.7 Sonnet and Claude Code". Anthropic. 24 February 2025. Retrieved 19 February 2026.
  55. ↑ "Upgraded Anthropic Console". anthropic.com. Retrieved 31 March 2025.
  56. ↑ "Token Saving Updates". anthropic.com. Retrieved 31 March 2025.
  57. ↑ "Web Search: Introducing Our Latest Advancement". anthropic.com. Retrieved 31 March 2025.
  58. ↑ "Introducing Claude 4". Anthropic. 22 May 2025. Retrieved 19 February 2026.
  59. ↑ "Activating AI Safety Level 3 protections". Anthropic. 22 May 2025. Retrieved 24 April 2026.
  60. ↑ "Anthropic is Quietly Backpedalling on its Safety Commitments". EA Forum. 23 May 2025. Retrieved 24 April 2026.
  61. ↑ "Anthropic's new AI model didn't just "blackmail" researchers in tests — it tried to leak information to news outlets". Nieman Lab. 27 May 2025. Retrieved 24 April 2026.
  62. ↑ "Anthropic Upgrades App-Building Capabilities to Claude Artifacts". InfoQ. 30 June 2025. Retrieved 25 September 2025.
  63. ↑ Robison, Kylie (2025-08-01). "Anthropic Revokes OpenAI's Access to Claude". Wired. Retrieved 2026-02-23.
  64. ↑ "Claude Opus 4.1". Anthropic. 5 August 2025. Retrieved 19 February 2026.
  65. ↑ Robison, Kylie (5 August 2025). "Claude Fans Threw a Funeral for Anthropic's Retired AI Model". Wired. Retrieved 23 February 2026.
  66. ↑ "Exclusive: How Claude became an emotional support bot". Axios. 26 June 2025. Retrieved 24 April 2026.
  67. ↑ "Anthropic Integrates AI Agent into Chrome". ForkLog. 27 August 2025. Retrieved 25 September 2025.
  68. ↑ "Introducing Claude Sonnet 4.5". Anthropic. 29 September 2025. Retrieved 19 February 2026.
  69. ↑ "Deloitte Accelerates Claude Expansion Amid Government Refund Controversy". The National CIO Review. 7 October 2025. Retrieved 20 February 2026.
  70. ↑ "Introducing Claude Haiku 4.5". Anthropic. 15 October 2025. Retrieved 18 February 2026.
  71. ↑ "Anthropic's Claude Takes Control of a Robot Dog". Wired. 12 November 2025. Retrieved 20 February 2026.
  72. ↑ Anthropic Team (November 24, 2025). "Introducing Claude Opus 4.5". Anthropic. Retrieved December 9, 2025.
  73. ↑ Bellan, Rebecca (December 8, 2025). "Claude Code is coming to Slack, and that's a bigger deal than it sounds". TechCrunch. Retrieved December 9, 2025.
  74. ↑ "NASA taps Claude to conjure Mars rover's travel plan". The Register. 31 January 2026. Retrieved 20 February 2026.
  75. ↑ "Google engineer says Claude Code built in one hour what her team spent a year on". The Decoder. 4 January 2026. Retrieved 20 February 2026.
  76. ↑ "Anthropic's Claude Cowork Is an AI Agent That Actually Works". Wired. 15 January 2026. Retrieved 20 February 2026.
  77. ↑ "Claude Cowork Is a Productivity Test Enterprises May Fail". Reworked. 18 February 2026. Retrieved 24 April 2026.
  78. ↑ "Claude's new constitution". Anthropic. 22 January 2026. Retrieved 19 February 2026.
  79. ↑ "Claude Code is suddenly everywhere inside Microsoft". The Verge. 22 January 2026. Retrieved 20 February 2026.
  80. ↑ Steinschaden, Jakob (3 February 2026). "Claude Cowork Triggers Tech Stock Selloff as AI Threatens SaaS Business Models". Trending Topics. Retrieved 23 February 2026.
  81. ↑ "Introducing Claude Opus 4.6". Anthropic. 5 February 2026. Retrieved 20 February 2026.
  82. ↑ "Introducing Claude Sonnet 4.6". Anthropic. 17 February 2026. Retrieved 20 February 2026.
  83. ↑ "Responsible Scaling Policy Version 3.0". Anthropic. 24 February 2026. Retrieved 24 April 2026.
  84. ↑ Claburn, Thomas (13 April 2026). "Claude is getting worse, according to Claude". The Register. Retrieved 20 April 2026.
  85. ↑ "Claude.ai down". Claude Status. 13 April 2026. Retrieved 20 April 2026.
  86. ↑ "Introducing Claude Opus 4.7". Anthropic. 16 April 2026. Retrieved 20 April 2026.
  87. ↑ Horsey, Julian (13 April 2026). "Claude Opus 4.7 Leaks & Anthropic's Full-Stack AI Studio". Geeky Gadgets. Retrieved 20 April 2026.
  88. ↑ "Anthropic releases Claude Opus 4.7 model as it previews more powerful Mythos AI". CNBC. 16 April 2026. Retrieved 20 April 2026.
  89. ↑ "Snowflake Adds Claude Opus 4.7 to AI Toolkit". StartupHub.ai. 16 April 2026. Retrieved 20 April 2026.
  90. ↑ "Announcing Claude Opus 4.7 on Snowflake Cortex AI". Snowflake. 16 April 2026. Retrieved 20 April 2026.
  91. ↑ "Snowflake and Anthropic announce $200 million partnership to bring agentic AI to global enterprises". Anthropic. 16 April 2026. Retrieved 20 April 2026.
  92. ↑ "Introducing Claude Design by Anthropic Labs". Anthropic. 17 April 2026. Retrieved 20 April 2026.
  93. ↑ "Interest in "Claude" over time". Google Trends. Google. Retrieved 22 February 2026.
  94. ↑ "Wikipedia pageviews for "Claude (language model)" (Jan 2023–Jan 2026)". Wikipedia Views. Retrieved 24 February 2026.
Retrieved from "https://timelines.issarice.com/index.php?title=Timeline_of_Claude&oldid=90224"