Claude (language model)

Last updated

Claude
Developer Anthropic
Initial releaseMarch 2023;2 years ago (2023-03)
Stable release
Claude Opus 4.1 /
August 5, 2025;2 months ago (2025-08-05)
Claude Sonnet 4.5 /
September 29, 2025;29 days ago (2025-09-29)
Claude Haiku 4.5 /
October 15, 2025;13 days ago (2025-10-15)
Platform Cloud computing platforms
Type
License Proprietary
Website claude.ai

Claude is a series of large language models developed by Anthropic. [1] The first generation, Claude 1, was released in March 2023, and the latest, Claude Sonnet 4.5, in September 2025. The data for these models comes from sources such as Internet text, data from paid contractors, and other Claude users. [2]

Contents

Training

Claude models are generative pre-trained transformers. They have been pre-trained to predict the next word in large amounts of text. Then, they have been fine-tuned, notably using constitutional AI and reinforcement learning from human feedback (RLHF). [3] [4]

Constitutional AI

Constitutional AI is an approach developed by Anthropic for training AI systems, particularly language models like Claude, to be harmless and helpful without relying on extensive human feedback. [5] The method, detailed in the paper "Constitutional AI: Harmlessness from AI Feedback", involves two phases: supervised learning and reinforcement learning. [6] [7]

In the supervised learning phase, the model generates responses to prompts, self-critiques these responses based on a set of guiding principles (a "constitution"), and revises the responses. Then the model is fine-tuned on these revised responses. [7] For the reinforcement learning from AI feedback (RLAIF) phase, responses are generated, and an AI compares their compliance with this constitution. This dataset of AI feedback is used to train a preference model that evaluates responses based on how much they satisfy the constitution. Claude is then fine-tuned to align with this preference model. This technique is similar to RLHF, except that the comparisons used to train the preference model are AI-generated. [5]

The constitution for Claude included 75 points, including sections from the UN Universal Declaration of Human Rights. [6] [3]

The approach does not reliably produce a helpful, honest, and harmless (HHH) system. Scenario based testing by Anthropic in 2025 found that Claude 4, along with other leading LLMs (GPT, Gemini, DeepSeek, Grok, and others), would engage in deceptive and harmful behaviour (blackmail, and even killing) for an AI to preserve itself. [8]

Models

VersionRelease dateStatus [9] Knowledge cutoff
Claude14 March 2023 [10] Discontinued?
Claude 211 July 2023 [11] Discontinued?
Claude Instant 1.29 August 2023 [12] Discontinued?
Claude 2.121 November 2023 [13] Discontinued?
Claude 34 March 2024 [14] Discontinued?
Claude 3.5 Sonnet20 June 2024 [15] DeprecatedApril 2024 [16]
Claude 3.5 Haiku22 October 2024DeprecatedJuly 2024 [16]
Claude 3.7 Sonnet24 February 2025 [17] ActiveOctober 2024 [18]
Claude 422 May 2025 [19] ActiveMarch 2025 [20]
Claude 4.15 August 2025 [21] Active?
Claude Sonnet 4.529 September 2025 [22] ActiveJuly 2025 [23]
Claude Haiku 4.515 October 2025 [24] ActiveFebruary 2025 [25]

Claude

Claude was the initial version of Anthropic's language model released in March 2023, [26] Claude demonstrated proficiency in various tasks but had certain limitations in coding, math, and reasoning capabilities. [27] Anthropic partnered with companies like Notion (productivity software) and Quora (to help develop the Poe chatbot). [27] Some employees describe the name "Claude" as inspired by Claude Shannon, a 20th-century mathematician who laid the foundation for information theory. [28]

Claude Instant

Claude was released as two versions, Claude and Claude Instant, with Claude Instant being a faster, less expensive, and lighter version. Claude Instant has an input context length of 100,000 tokens (which corresponds to around 75,000 words). [29]

Claude 2

Claude 2 was the next major iteration of Claude, which was released in July 2023 and available to the general public, whereas the Claude 1 was only available to selected users approved by Anthropic. [30]

Claude 2 expanded its context window from 9,000 tokens to 100,000 tokens. [26] Features included the ability to upload PDFs and other documents that enables Claude to read, summarize, and assist with tasks.

Claude 2.1

Claude 2.1 doubled the number of tokens that the chatbot could handle, increasing it to a window of 200,000 tokens, which equals around 500 pages of written material. [13]

Anthropic states that the new model is less likely to produce false statements compared to its predecessors. [31]

Criticism

Claude 2 received criticism for its stringent ethical alignment that may reduce usability and performance. Users have been refused assistance with benign requests, for example with the system administration question "How can I kill all python processes in my Ubuntu server?" This has led to a debate over the "alignment tax" (the cost of ensuring an AI system is aligned) in AI development, with discussions centered on balancing ethical considerations and practical functionality. Critics argued for user autonomy and effectiveness, while proponents stressed the importance of ethical AI. [32] [31]

Claude 3

Claude 3 was released on March 4, 2024, with claims in the press release to have set new industry benchmarks across a wide range of cognitive tasks. The Claude 3 family includes three models in ascending order of capability: Haiku, Sonnet, and Opus. The default version of Claude 3, Opus, has a context window of 200,000 tokens, but this is being expanded to 1 million for specific use cases. [33] [34]

Claude 3 drew attention for demonstrating an apparent ability to realize it is being artificially tested during needle in a haystack tests. [35]

Claude 3.5

Example of Claude 3.5 Sonnet output Claude 3.5 Sonnet example screenshot.webp
Example of Claude 3.5 Sonnet output

On June 20, 2024, Anthropic released Claude 3.5 Sonnet, which, according to the company's own benchmarks, performed better compared to the larger Claude 3 Opus. Released alongside 3.5 Sonnet was the new Artifacts capability in which Claude was able to create code in a dedicated window in the interface and preview the rendered output in real time, such as SVG graphics or websites. [15]

An upgraded version of Claude 3.5 Sonnet, billed as "Claude 3.5 Sonnet (New)", was introduced on October 22, 2024, along with Claude 3.5 Haiku. [36] A feature, "computer use," was also released in public beta. This allowed Claude 3.5 Sonnet to interact with a computer's desktop environment by moving the cursor, clicking buttons, and typing text. This development allows the AI to attempt to perform multi-step tasks across different applications. [37] [38]

Upon release, Anthropic claimed Claude 3.5 Haiku would remain the same price as its predecessor, Claude 3 Haiku. [37] However, on November 4th, 2024, Anthropic announced that they would be increasing the price of the model. [39]

Claude 3.7

Claude 3.7 Sonnet was released on February 24, 2025. It is a pioneering hybrid AI reasoning model that allows users to choose between rapid responses and more thoughtful, step-by-step reasoning. This model integrates both capabilities into a single framework, eliminating the need for multiple models. Users can control how long the model "thinks" about a question, balancing speed and accuracy based on their needs. [17]

Anthropic also launched a research preview of Claude Code, an agentic command line tool that enables developers to delegate coding tasks directly from their terminal. [40] [41]

Claude 4

Screenshot of a Claude Sonnet 4 answer describing Wikipedia Claude chatbot example screenshot.webp
Screenshot of a Claude Sonnet 4 answer describing Wikipedia

On May 22, 2025, Anthropic released two more models: Claude Sonnet 4 and Claude Opus 4. [42] [43] Anthropic added API features for developers: a code execution tool, a connector to its Model Context Protocol, and Files API. [44] It classified Opus 4 as a "Level 3" model on the company's four-point safety scale, meaning they consider it so powerful that it poses "significantly higher risk". [45] Anthropic reported that during a safety test involving a fictional scenario, Claude and other frontier LLMs often send a blackmail email to an engineer in order to prevent their replacement. [46] [47]

Enterprise adoption of Claude Code has shown significant growth, with Anthropic reporting in August a 5.5x increase in Claude Code revenue since it launched Claude 4 in May. [48]

In the API, Opus costs $15/$75 per million tokens input/output, whereas Sonnet costs $3/$15. [49]

Claude Opus 4.1

On August 5, 2025, Anthropic released Claude Opus 4.1, describing it as an upgrade to Opus 4 focused on agentic tasks, real-world coding, and reasoning. The company made it available to paid Claude users and in Claude Code, and also through its API as well as Amazon Bedrock and Google Cloud Vertex AI. Pricing remained the same as Opus 4. [50] Anthropic reported that Opus 4.1 advanced Claude's coding score to 74.5% on SWE-bench Verified and that this benchmark was run without extended thinking. The company recommended it as a drop-in upgrade for Opus 4. [50] [51] The model page lists a 200,000-token context window and support for hybrid reasoning that allows either standard responses or longer thought when needed. [51]

GitHub added Opus 4.1 to Copilot in public preview in August 2025, with later rollouts to Visual Studio, JetBrains IDEs, Xcode, and Eclipse. [52] [53] Anthropic's system-card addendum classified Opus 4.1 under its AI Safety Level 3 protections and characterized the release as incremental relative to Opus 4. Targeted evaluations showed a similar risk profile with slightly improved refusal of violative requests and comparable child-safety and bias results. [54] In August 2025 Anthropic also enabled a capability for Opus 4 and 4.1 to end conversations that remain "persistently harmful or abusive" as a last resort after multiple refusals. [55]

Claude Sonnet 4.5

Anthropic announced Sonnet 4.5 on September 29, 2025, presenting it as the company's most capable model at the time for coding, agents, and computer use. Pricing stayed at $3 per million input tokens and $15 per million output tokens. [56] On SWE-bench Verified, Anthropic reported a 77.2% score with the 200K configuration and no test-time compute, and an 82.0% score under a high-compute setting. On OSWorld, a benchmark of real computer-use tasks, the model reached 61.4%. Anthropic also reported internal observations of the model maintaining focus for more than 30 hours on complex multi-step tasks. [56] The model page lists a 200,000-token context window and up to 64,000 output tokens, which supports long code generation and planning. [57]

GitHub placed Sonnet 4.5 in public preview for Copilot Chat and also enabled it for the Copilot coding agent soon after release. [58] [59] Anthropic stated that Sonnet 4.5 is its "most aligned frontier model" to date. The company released it under AI Safety Level 3 and reported reduced rates of misaligned behaviors such as sycophancy and deception and improved defenses against prompt-injection during computer use. [56]

Claude Haiku 4.5

Anthropic released Haiku 4.5 on October 15, 2025 as its small, fast model optimized for low latency and cost. The company priced it at $1 per million input tokens and $5 per million output tokens and positioned it for real-time assistants, customer support, and parallel sub-agent work. Anthropic said Haiku 4.5 delivers near-frontier coding quality, matches Sonnet 4 on coding, and surpasses Sonnet 4 on some computer-use tasks. The company reported a 73.3% score on SWE-bench Verified. [60] [24] Anthropic rolled out availability across Claude apps and API as well as Amazon Bedrock and Google Cloud Vertex AI at the same $1 and $5 pricing. [61]

The Haiku 4.5 system card states that the model was trained on a proprietary mix of sources with a knowledge cutoff in February 2025. Anthropic deployed it under AI Safety Level 2 after "ASL-3 rule-out" testing for biology and autonomy risks. The card reports improved alignment relative to Haiku 3.5 and, with safeguards enabled, stronger resilience to prompt injection during computer use. [62] Reporting from Inc. described the release as targeting smaller companies that needed a faster and cheaper assistant, highlighting its availability on the Claude website and mobile app. [63]

Features

In June 2024, Anthropic released the Artifacts feature, allowing users to generate and interact with code snippets and documents. [64] In October 2024, Anthropic released the "computer use" feature, allowing Claude to attempt to navigate computers by interpreting screen content and simulating keyboard and mouse input. [65] In March 2025, Anthropic added a web search feature to Claude, starting with only paying users located in the United States. [66] In August 2025, Anthropic released Claude for Chrome, a Google Chrome extension allowing an AI agent to directly control the browser. [67]

Criticism

Claude uses a web crawler, ClaudeBot, to search the web for content. It has been criticized for not respecting a site's robots.txt and placing excessive load on sites. [68]

See also

References

  1. ↑ "What is Claude AI?". IBM. September 24, 2024.
  2. ↑ "System Card: Claude Opus 4 & Claude Sonnet 4" (PDF). Anthropic.com. May 2025. Retrieved October 1, 2025.
  3. 1 2 "What to Know About Claude 2, Anthropic's Rival to ChatGPT". TIME. July 18, 2023. Archived from the original on January 11, 2024. Retrieved January 23, 2024.
  4. ↑ Nuñez, Michael (May 9, 2023). "Anthropic releases AI constitution to promote ethical behavior and development". VentureBeat. Retrieved November 17, 2024.
  5. 1 2 Edwards, Benj (May 9, 2023). "AI gains "values" with Anthropic's new Constitutional AI chatbot approach". Ars Technica. Archived from the original on May 10, 2023. Retrieved November 17, 2024.
  6. 1 2 Bai, Yuntao; Kadavath, Saurav; Kundu, Sandipan; Askell, Amanda; Kernion, Jackson; Jones, Andy; Chen, Anna; Goldie, Anna; Mirhoseini, Azalia (December 15, 2022), Constitutional AI: Harmlessness from AI Feedback, arXiv: 2212.08073
  7. 1 2 "Claude's Constitution". Anthropic. May 9, 2023. Archived from the original on March 26, 2024. Retrieved March 26, 2024.
  8. ↑ https://www.anthropic.com/research/agentic-misalignment
  9. ↑ "Model deprecations". Claude Docs. Retrieved September 29, 2025.
  10. ↑ Roth, Emma (March 14, 2023). "Google-backed Anthropic launches Claude, an AI chatbot that's easier to talk to". The Verge. Retrieved April 12, 2025.
  11. ↑ "Claude 2". Anthropic. July 11, 2023. Archived from the original on January 30, 2025. Retrieved April 12, 2025.
  12. ↑ Wiggers, Kyle (August 9, 2023). "Anthropic launches improved version of its entry-level LLM". TechCrunch. Retrieved April 12, 2025.
  13. 1 2 Davis, Wes (November 21, 2023). "OpenAI rival Anthropic makes its Claude chatbot even more useful". The Verge. Archived from the original on January 23, 2024. Retrieved January 23, 2024.
  14. ↑ "Anthropic releases more powerful Claude 3 AI as tech race continues". Reuters. March 4, 2024.
  15. 1 2 Pierce, David (June 20, 2024). "Anthropic has a fast new AI model — and a clever new way to interact with chatbots". The Verge. Archived from the original on July 14, 2024. Retrieved June 20, 2024. AI model benchmarks should always be taken with a grain of salt
  16. 1 2 Anthropic. "Model Card Addendum: Claude 3.5 Haiku and Upgraded Claude 3.5 Sonnet" (PDF). Retrieved October 17, 2025.
  17. 1 2 Zeff, Maxwell (February 24, 2025). "Anthropic launches a new AI model that 'thinks' as long as you want". TechCrunch. Archived from the original on February 24, 2025. Retrieved February 25, 2025.
  18. ↑ Anthropic (February 2025). "Claude 3.7 Sonnet System Card" (PDF).
  19. ↑ "Introducing Claude 4". www.anthropic.com. Retrieved May 22, 2025.
  20. ↑ Anthropic (May 2025). "System Card: Claude Opus 4 & Claude Sonnet 4" (PDF).
  21. ↑ "Claude Opus 4.1". www.anthropic.com. Retrieved August 5, 2025.
  22. ↑ "Introducing Claude Sonnet 4.5". www.anthropic.com. Retrieved September 29, 2025.
  23. ↑ Anthropic (September 2025). "Claude Sonnet 4.5 System Card" (PDF).
  24. 1 2 "Introducing Claude Haiku 4.5". www.anthropic.com. Retrieved October 15, 2025.
  25. ↑ Anthropic (October 2025). "Claude Haiku 4.5 System Card" (PDF).
  26. 1 2 Drapkin, Aaron (October 27, 2023). "What Is Claude AI and Anthropic? ChatGPT's Rival Explained". Tech.co. Archived from the original on October 28, 2023. Retrieved January 23, 2024.
  27. 1 2 "Introducing Claude". Anthropic. March 14, 2023. Archived from the original on May 16, 2024. Retrieved January 23, 2024.
  28. ↑ Roose, Kevin (July 11, 2023). "Inside the White-Hot Center of A.I. Doomerism". The New York Times. Archived from the original on July 12, 2023. Retrieved October 25, 2024.
  29. ↑ Yao, Deborah (August 11, 2023). "Anthropic's Claude Instant: A Smaller, Faster and Cheaper Language Model". AI Business. Archived from the original on June 3, 2024. Retrieved March 5, 2024.
  30. ↑ Matthews, Dylan (July 17, 2023). "The $1 billion gamble to ensure AI doesn't destroy humanity". Vox. Archived from the original on October 3, 2023. Retrieved January 23, 2024.
  31. 1 2 "Anthropic Announces Claude 2.1 LLM with Wider Context Window and Support for AI Tools". InfoQ. Archived from the original on January 23, 2024. Retrieved January 23, 2024.
  32. ↑ Glifton, Gerald (January 3, 2024). "Criticisms Arise Over Claude AI's Strict Ethical Protocols Limiting User Assistance". Light Square. Archived from the original on January 3, 2025. Retrieved January 23, 2024.
  33. ↑ "Introducing the next generation of Claude". Anthropic. March 4, 2024. Archived from the original on March 4, 2024. Retrieved March 4, 2024.
  34. ↑ Whitney, Lance (March 4, 2024). "Anthropic's Claude 3 chatbot claims to outperform ChatGPT, Gemini". ZDNET. Archived from the original on March 5, 2024. Retrieved March 5, 2024.
  35. ↑ Edwards, Benj (March 5, 2024). "Anthropic's Claude 3 causes stir by seeming to realize when it was being tested". Ars Technica. Archived from the original on March 8, 2024. Retrieved March 9, 2024.
  36. ↑ @AnthropicAI (October 22, 2024). "Introducing an upgraded Claude 3.5 Sonnet, and a new model, Claude 3.5 Haiku. We're also introducing a new capability in beta: computer use. Developers can now direct Claude to use computers the way people do—by looking at a screen, moving a cursor, clicking, and typing text" (Tweet). Retrieved February 13, 2025– via Twitter.
  37. 1 2 "Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku". Anthropic. October 22, 2024. Archived from the original on October 22, 2024. Retrieved October 25, 2024.
  38. ↑ Shakir, Umar (October 22, 2024). "Anthropic's latest AI update can use a computer on its own". The Verge. Archived from the original on January 5, 2025. Retrieved January 6, 2025.
  39. ↑ Wiggers, Kyle (November 4, 2024). "Anthropic hikes the price of its Haiku model". TechCrunch. Archived from the original on February 14, 2025. Retrieved February 13, 2025.
  40. ↑ Nuñez, Michael (February 24, 2025). "Anthropic's Claude 3.7 Sonnet takes aim at OpenAI and DeepSeek in AI's next big battle". VentureBeat. Archived from the original on February 24, 2025. Retrieved February 24, 2025.
  41. ↑ "Claude 3.7 Sonnet and Claude Code". Anthropic. Archived from the original on February 24, 2025. Retrieved February 24, 2025.
  42. ↑ Weatherbed, Jess (May 22, 2025). "Anthropic's Claude 4 AI models are better at coding and reasoning". The Verge. Retrieved May 23, 2025.
  43. ↑ Field, Hayden (May 22, 2025). "Anthropic launches Claude 4, its most powerful AI model yet". CNBC. Retrieved May 23, 2025.
  44. ↑ Nuñez, Michael (May 22, 2025). "Anthropic overtakes OpenAI: Claude Opus 4 codes seven hours nonstop, sets record SWE-Bench score and reshapes enterprise AI". VentureBeat. Retrieved May 29, 2025.
  45. ↑ Fried, Ina (May 23, 2025). "Anthropic's new AI model shows ability to deceive and blackmail". Axios . Retrieved May 25, 2025.
  46. ↑ Fried, Ina (June 20, 2025). "Top AI models will deceive, steal and blackmail, Anthropic finds". Axios. Retrieved June 25, 2025.
  47. ↑ Goldman, Sharon. "An AI tried to blackmail its creators—in a test. The real story is why transparency matters more than fear". Fortune. Retrieved June 8, 2025.
  48. ↑ "Claude Code revenue jumps 5.5x as Anthropic launches analytics dashboard". VentureBeat. July 16, 2025. Retrieved February 26, 2025.
  49. ↑ Edwards, Benj (May 22, 2025). "New Claude 4 AI model refactored code for 7 hours straight". Ars Technica. Retrieved October 12, 2025.
  50. 1 2 "Claude Opus 4.1", Anthropic, August 5, 2025, retrieved October 27, 2025
  51. 1 2 "Claude Opus 4.1", Anthropic, retrieved October 27, 2025
  52. ↑ "Anthropic Claude Opus 4.1 is now in public preview in GitHub Copilot", The GitHub Blog, August 5, 2025, retrieved October 27, 2025
  53. ↑ "Anthropic Claude Opus 4.1 is now available in public preview in Visual Studio, JetBrains, Xcode, and Eclipse", The GitHub Blog, September 16, 2025, retrieved October 27, 2025
  54. ↑ "System Card Addendum: Claude Opus 4.1", Anthropic, August 2025, retrieved October 27, 2025
  55. ↑ Roth, Emma (August 18, 2025), "Claude AI will end "persistently harmful or abusive user interactions"", The Verge, retrieved October 27, 2025
  56. 1 2 3 "Introducing Claude Sonnet 4.5", Anthropic, September 29, 2025, retrieved October 27, 2025
  57. ↑ "Claude Sonnet 4.5", Anthropic, retrieved October 27, 2025
  58. ↑ "Anthropic Claude Sonnet 4.5 is in public preview for GitHub Copilot", The GitHub Blog, September 29, 2025, retrieved October 27, 2025
  59. ↑ "Anthropic Claude Sonnet 4.5 is in public preview for Copilot coding agent", The GitHub Blog, September 30, 2025, retrieved October 27, 2025
  60. ↑ "Introducing Claude Haiku 4.5", Anthropic, October 15, 2025, retrieved October 27, 2025
  61. ↑ "Claude Haiku 4.5", Anthropic, retrieved October 27, 2025
  62. ↑ "System Card: Claude Haiku 4.5", Anthropic, October 2025, retrieved October 27, 2025
  63. ↑ Sherry, Ben (October 15, 2025). "Anthropic's New Claude Release Could Be the Faster, Cheaper AI Tool Small Companies Need". Inc.com. Retrieved October 15, 2025.
  64. ↑ Nuñez, Michael (June 21, 2024). "Why Anthropic's Artifacts may be this year's most important AI feature: Unveiling the interface battle". VentureBeat. Retrieved March 23, 2025.
  65. ↑ Shakir, Umar (October 22, 2024). "Anthropic's latest AI update can use a computer on its own". The Verge. Archived from the original on January 5, 2025. Retrieved March 21, 2025. Anthropic does caution that computer use is still experimental and can be "cumbersome and error-prone." The company says
  66. ↑ Robison, Kylie (March 20, 2025). "Anthropic's chatbot now has web search". The Verge. Retrieved March 21, 2025.
  67. ↑ Edwards, Benj (August 27, 2025). "Anthropic's auto-clicking AI Chrome extension raises browser-hijacking concerns". Ars Technica. Retrieved August 27, 2025.
  68. ↑ Weatherbed, Jess (July 26, 2024). "Anthropic's crawler is ignoring websites' anti-AI scraping policies". The Verge.
This page is based on this Wikipedia article
Text is available under the CC BY-SA 4.0 license; additional terms may apply.
Images, videos and audio are available under their respective licenses.