Claude (language model)

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).{{Cite web |last=Nuñez |first=Michael |date=2023-05-09 |title=Anthropic releases AI constitution to promote ethical behavior and development |url=https://venturebeat.com/ai/anthropic-releases-ai-constitution-to-promote-ethical-behavior-and-development/ |access-date=2024-11-17 |website=VentureBeat |language=en-US}}

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.{{Cite web |last=Edwards |first=Benj |date=2023-05-09 |title=AI gains "values" with Anthropic's new Constitutional AI chatbot approach |url=https://arstechnica.com/information-technology/2023/05/ai-with-a-moral-compass-anthropic-outlines-constitutional-ai-in-its-claude-chatbot/ |access-date=2024-11-17 |website=Ars Technica |language=en-US |archive-date=May 10, 2023 |archive-url=https://archive.today/20230510111936/https://arstechnica.com/information-technology/2023/05/ai-with-a-moral-compass-anthropic-outlines-constitutional-ai-in-its-claude-chatbot/ |url-status=live }} The method, detailed in the paper "Constitutional AI: Harmlessness from AI Feedback" involves two phases: supervised learning and reinforcement learning.{{Citation |last1=Bai |first1=Yuntao |title=Constitutional AI: Harmlessness from AI Feedback |date=December 15, 2022 |arxiv=2212.08073 |last2=Kadavath |first2=Saurav |last3=Kundu |first3=Sandipan |last4=Askell |first4=Amanda |last5=Kernion |first5=Jackson |last6=Jones |first6=Andy |last7=Chen |first7=Anna |last8=Goldie |first8=Anna |last9=Mirhoseini |first9=Azalia}}{{Cite web |date=May 9, 2023 |title=Claude's Constitution |url=https://www.anthropic.com/news/claudes-constitution |access-date=March 26, 2024 |website=Anthropic |language=en |archive-date=March 26, 2024 |archive-url=https://web.archive.org/web/20240326110757/https://www.anthropic.com/news/claudes-constitution |url-status=live }}

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. 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.

The constitution for Claude included 75 points, including sections from the UN Universal Declaration of Human Rights.{{Cite magazine |date=July 18, 2023 |title=What to Know About Claude 2, Anthropic's Rival to ChatGPT |url=https://time.com/6295523/claude-2-anthropic-chatgpt/ |access-date=January 23, 2024 |magazine=TIME |language=en |archive-date=January 11, 2024 |archive-url=https://web.archive.org/web/20240111192116/https://time.com/6295523/claude-2-anthropic-chatgpt/ |url-status=live }}

File:Claude chatbot example screenshot.webp]]

Claude is named after Claude Shannon, a pioneer in AI research.{{Cite news |last=Roose |first=Kevin |date=July 11, 2023 |title=Inside the White-Hot Center of A.I. Doomerism |url=https://www.nytimes.com/2023/07/11/technology/anthropic-ai-claude-chatbot.html |work=The New York Times |archive-date=July 12, 2023 |access-date=October 25, 2024 |archive-url=https://web.archive.org/web/20230712235416/https://www.nytimes.com/2023/07/11/technology/anthropic-ai-claude-chatbot.html |url-status=live }}

Claude was the initial version of Anthropic's language model released in March 2023,{{Cite web |last=Drapkin |first=Aaron |date=October 27, 2023 |title=What Is Claude AI and Anthropic? ChatGPT's Rival Explained |url=https://tech.co/news/what-is-claude-ai-anthropic |access-date=January 23, 2024 |website=Tech.co |language=en-US |archive-date=October 28, 2023 |archive-url=https://web.archive.org/web/20231028051551/https://tech.co/news/what-is-claude-ai-anthropic |url-status=live }} Claude demonstrated proficiency in various tasks but had certain limitations in coding, math, and reasoning capabilities.{{Cite web |date=March 14, 2023 |title=Introducing Claude |url=https://www.anthropic.com/news/introducing-claude |website=Anthropic |access-date=January 23, 2024 |archive-date=May 16, 2024 |archive-url=https://web.archive.org/web/20240516063358/https://www.anthropic.com/news/introducing-claude |url-status=live }} Anthropic partnered with companies like Notion (productivity software) and Quora (to help develop the Poe chatbot).

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).{{Cite news |last=Yao |first=Deborah |date=August 11, 2023 |title=Anthropic's Claude Instant: A Smaller, Faster and Cheaper Language Model |url=https://aibusiness.com/nlp/anthropic-s-claude-instant-a-smaller-faster-and-cheaper-language-model |work=AI Business |archive-date=June 3, 2024 |access-date=March 5, 2024 |archive-url=https://web.archive.org/web/20240603171639/https://aibusiness.com/nlp/anthropic-s-claude-instant-a-smaller-faster-and-cheaper-language-model |url-status=live }}

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.{{Cite web |last=Matthews |first=Dylan |date=July 17, 2023 |title=The $1 billion gamble to ensure AI doesn't destroy humanity |url=https://www.vox.com/future-perfect/23794855/anthropic-ai-openai-claude-2 |access-date=January 23, 2024 |website=Vox |language=en |archive-date=October 3, 2023 |archive-url=https://web.archive.org/web/20231003205441/https://www.vox.com/future-perfect/23794855/anthropic-ai-openai-claude-2 |url-status=live }}

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

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.{{Cite web |last=Davis |first=Wes |date=November 21, 2023 |title=OpenAI rival Anthropic makes its Claude chatbot even more useful |url=https://www.theverge.com/2023/11/21/23971070/anthropic-claude-2-1-openai-ai-chatbot-update-beta-tools |access-date=January 23, 2024 |website=The Verge |language=en |archive-date=January 23, 2024 |archive-url=https://web.archive.org/web/20240123003015/https://www.theverge.com/2023/11/21/23971070/anthropic-claude-2-1-openai-ai-chatbot-update-beta-tools |url-status=live }}

Anthropic states that the new model is less likely to produce false statements compared to its predecessors.{{Cite web |title=Anthropic Announces Claude 2.1 LLM with Wider Context Window and Support for AI Tools |url=https://www.infoq.com/news/2023/11/anthropic-announces-claude-2-1/ |access-date=January 23, 2024 |website=InfoQ |language=en |archive-date=January 23, 2024 |archive-url=https://web.archive.org/web/20240123004145/https://www.infoq.com/news/2023/11/anthropic-announces-claude-2-1/ |url-status=live }}

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.{{Cite web |last=Glifton |first=Gerald |date=January 3, 2024 |title=Criticisms Arise Over Claude AI's Strict Ethical Protocols Limiting User Assistance |url=https://lightsquare.org/news/criticisms-arise-over-claude-ais-strict-ethical-protocols-limiting-user-assistance |access-date=January 23, 2024 |website=Light Square |language=en |archive-date=January 3, 2025 |archive-url=https://web.archive.org/web/20250103170739/https://lightsquare.org/news/criticisms-arise-over-claude-ais-strict-ethical-protocols-limiting-user-assistance |url-status=live }}

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 state-of-the-art 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.{{Cite web |date=2024-03-04 |title=Introducing the next generation of Claude |url=https://www.anthropic.com/news/claude-3-family |access-date=March 4, 2024 |website=Anthropic |language=en |archive-date=March 4, 2024 |archive-url=https://web.archive.org/web/20240304143650/https://www.anthropic.com/news/claude-3-family |url-status=live }}{{Cite web |last=Whitney |first=Lance |date=March 4, 2024 |title=Anthropic's Claude 3 chatbot claims to outperform ChatGPT, Gemini |url=https://www.zdnet.com/article/anthropics-claude-3-chatbot-claims-to-outperform-chatgpt-gemini/ |access-date=March 5, 2024 |website=ZDNET |language=en |archive-date=March 5, 2024 |archive-url=https://web.archive.org/web/20240305024829/https://www.zdnet.com/article/anthropics-claude-3-chatbot-claims-to-outperform-chatgpt-gemini/ |url-status=live }}

Claude 3 drew attention for demonstrating an apparent ability to realize it is being artificially tested during needle in a haystack tests.{{Cite web |last=Edwards |first=Benj |date=March 5, 2024 |title=Anthropic's Claude 3 causes stir by seeming to realize when it was being tested |url=https://arstechnica.com/information-technology/2024/03/claude-3-seems-to-detect-when-it-is-being-tested-sparking-ai-buzz-online/ |access-date=March 9, 2024 |website=Ars Technica |language=en-us |archive-date=March 8, 2024 |archive-url=https://web.archive.org/web/20240308165441/https://arstechnica.com/information-technology/2024/03/claude-3-seems-to-detect-when-it-is-being-tested-sparking-ai-buzz-online/ |url-status=live }}

File:Claude 3.5 Sonnet example screenshot.webp

On June 20, 2024, Anthropic released Claude 3.5 Sonnet, which demonstrated significantly improved performance on benchmarks compared to the larger Claude 3 Opus, notably in areas such as coding, multistep workflows, chart interpretation, and text extraction from images. 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.{{Cite web |last=Pierce |first=David |date=June 20, 2024 |title=Anthropic has a fast new AI model — and a clever new way to interact with chatbots |url=https://www.theverge.com/2024/6/20/24181961/anthropic-claude-35-sonnet-model-ai-launch |access-date=June 20, 2024 |website=The Verge |language=en |archive-date=July 14, 2024 |archive-url=https://archive.today/20240714062712/https://www.theverge.com/2024/6/20/24181961/anthropic-claude-35-sonnet-model-ai-launch |url-status=live }} Anthropic also announced that Claude 3.5 Opus would be released later that year, and added it to their models page.{{Cite web |title=Learn about Claude - Models |url=https://docs.anthropic.com/en/docs/about-claude/models |archive-url=https://web.archive.org/web/20240625030318/https://docs.anthropic.com/en/docs/about-claude/models |archive-date=June 25, 2024 |access-date=February 13, 2025 |website=Anthropic Documentation}} However, as of February 2025, Claude 3.5 Opus has not been released, and Anthropic has removed mention of it from the models page.{{Cite web |title=About Claude Models |url=https://docs.anthropic.com/en/docs/about-claude/models |access-date=February 13, 2025 |website=Anthropic Documentation |archive-date=June 25, 2024 |archive-url=https://web.archive.org/web/20240625030318/https://docs.anthropic.com/en/docs/about-claude/models |url-status=live }}

An "upgraded Claude 3.5 Sonnet", billed as "Claude 3.5 Sonnet (New)" in the web interface and benchmarks, was introduced on October 22, 2024, along with Claude 3.5 Haiku.{{Cite tweet |number=1848742740420341988 |user=AnthropicAI |title=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. |date=October 22, 2024 |access-date=February 13, 2025 |url=https://x.com/AnthropicAI/status/1848742740420341988}} A feature, "computer use," was also unveiled in public beta. This capability enables Claude 3.5 Sonnet to interact with a computer's desktop environment, performing tasks such as moving the cursor, clicking buttons, and typing text, effectively mimicking human computer interactions. This development allows the AI to autonomously execute complex, multi-step tasks across various applications.{{Cite web |date=2024-10-22 |title=Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku |url=https://www.anthropic.com/news/3-5-models-and-computer-use |access-date=2024-10-25 |website=Anthropic |language=en |archive-date=October 22, 2024 |archive-url=https://web.archive.org/web/20241022151101/https://www.anthropic.com/news/3-5-models-and-computer-use |url-status=live }}{{Cite news |last=Shakir |first=Umar |date=October 22, 2024 |title=Anthropic's latest AI update can use a computer on its own |url=https://www.theverge.com/2024/10/22/24276822/anthopic-claude-3-5-sonnet-computer-use-ai |access-date=January 6, 2025 |work=The Verge |archive-date=January 5, 2025 |archive-url=https://web.archive.org/web/20250105154321/https://www.theverge.com/2024/10/22/24276822/anthopic-claude-3-5-sonnet-computer-use-ai |url-status=live }}

Upon release, Anthropic claimed Claude 3.5 Haiku would remain the same price as its predecessor, Claude 3 Haiku. However, on November 4th, 2024, Anthropic announced that they would be increasing the price of the model "to reflect its increase in intelligence".{{Cite news |last=Wiggers |first=Kyle |date=November 4, 2024 |title=Anthropic hikes the price of its Haiku model |url=https://techcrunch.com/2024/11/04/anthropic-hikes-the-price-of-its-haiku-model/ |access-date=February 13, 2025 |work=TechCrunch |archive-date=February 14, 2025 |archive-url=https://web.archive.org/web/20250214181406/https://techcrunch.com/2024/11/04/anthropic-hikes-the-price-of-its-haiku-model/ |url-status=live }}{{Cite tweet |number=1853498270724542658 |user=AnthropicAI |title=During final testing, Haiku surpassed Claude 3 Opus, our previous flagship model, on many benchmarks—at a fraction of the cost. As a result, we've increased pricing for Claude 3.5 Haiku to reflect its increase in intelligence. |date=November 4, 2024 |access-date=February 13, 2025 |url=https://x.com/AnthropicAI/status/1853498270724542658}}

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.{{Cite web |last=Zeff |first=Maxwell |date=2025-02-24 |title=Anthropic launches a new AI model that 'thinks' as long as you want |url=https://techcrunch.com/2025/02/24/anthropic-launches-a-new-ai-model-that-thinks-as-long-as-you-want/ |access-date=2025-02-25 |website=TechCrunch |language=en-US |archive-date=February 24, 2025 |archive-url=https://web.archive.org/web/20250224183412/https://techcrunch.com/2025/02/24/anthropic-launches-a-new-ai-model-that-thinks-as-long-as-you-want/ |url-status=live }}

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.{{Cite web |last=Nuñez |first=Michael |date=2025-02-24 |title=Anthropic's Claude 3.7 Sonnet takes aim at OpenAI and DeepSeek in AI's next big battle |url=https://venturebeat.com/ai/anthropics-claude-3-7-sonnet-takes-aim-at-openai-and-deepseek-in-ais-next-big-battle/ |access-date=2025-02-24 |website=VentureBeat |language=en-US |archive-date=February 24, 2025 |archive-url=https://web.archive.org/web/20250224183545/https://venturebeat.com/ai/anthropics-claude-3-7-sonnet-takes-aim-at-openai-and-deepseek-in-ais-next-big-battle/ |url-status=live }}{{Cite web |title=Claude 3.7 Sonnet and Claude Code |url=https://www.anthropic.com/news/claude-3-7-sonnet |access-date=February 24, 2025 |website=Anthropic |language=en |archive-date=February 24, 2025 |archive-url=https://web.archive.org/web/20250224183042/https://www.anthropic.com/news/claude-3-7-sonnet |url-status=live }}

On May 22, 2025, Anthropic released two more models: Claude Sonnet 4 and Claude Opus 4.{{Cite web |last=Weatherbed |first=Jess |date=2025-05-22 |title=Anthropic’s Claude 4 AI models are better at coding and reasoning |url=https://www.theverge.com/news/672705/anthropic-claude-4-ai-ous-sonnet-availability |access-date=2025-05-23 |website=The Verge |language=en-US}}{{Cite web |last=Field |first=Hayden |date=2025-05-22 |title=Anthropic launches Claude 4, its most powerful AI model yet |url=https://www.cnbc.com/2025/05/22/claude-4-opus-sonnet-anthropic.html |access-date=2025-05-23 |website=CNBC |language=en}} Anthropic added API features for developers: a code execution tool, a connector to its Model Context Protocol, and Files API.{{Cite web |last=Nuñez |first=Michael |date=2025-05-22 |title=Anthropic overtakes OpenAI: Claude Opus 4 codes seven hours nonstop, sets record SWE-Bench score and reshapes enterprise AI |url=https://venturebeat.com/ai/anthropic-claude-opus-4-can-code-for-7-hours-straight-and-its-about-to-change-how-we-work-with-ai/ |access-date=2025-05-29 |website=VentureBeat |language=en-US}} 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."{{Cite web |last=Fried |first=Ina |date=2025-05-23 |title=Anthropic's new AI model shows ability to deceive and blackmail |url=https://www.axios.com/2025/05/23/anthropic-ai-deception-risk |access-date=2025-05-25 |website=Axios |language=en}}

In June 2024, Anthropic released the Artifacts feature, allowing users to generate and interact with code snippets and documents.{{Cite web |last=Nuñez |first=Michael |date=2024-06-21 |title=Why Anthropic's Artifacts may be this year's most important AI feature: Unveiling the interface battle |url=https://venturebeat.com/ai/why-anthropics-artifacts-may-be-this-years-most-important-ai-feature-unveiling-the-interface-battle/ |access-date=2025-03-23 |website=VentureBeat |language=en-US}}

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.{{Cite web |last=Shakir |first=Umar |date=2024-10-22 |title=Anthropic's latest AI update can use a computer on its own |url=https://www.theverge.com/2024/10/22/24276822/anthopic-claude-3-5-sonnet-computer-use-ai |access-date=2025-03-21 |website=The Verge |language=en-US |quote=Anthropic does caution that computer use is still experimental and can be “cumbersome and error-prone.” The company says |archive-date=January 5, 2025 |archive-url=https://web.archive.org/web/20250105154321/https://www.theverge.com/2024/10/22/24276822/anthopic-claude-3-5-sonnet-computer-use-ai |url-status=live }}

In March 2025, Anthropic added a web search feature to Claude, starting with only paying users located in the United States.{{Cite web |last=Robison |first=Kylie |date=2025-03-20 |title=Anthropic's chatbot now has web search |url=https://www.theverge.com/ai-artificial-intelligence/632824/anthropic-claude-web-search-ai-scott-white |access-date=2025-03-21 |website=The Verge |language=en-US}}

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.{{Cite web |last=Weatherbed |first=Jess |date=2024-07-26 |url=https://www.theverge.com/2024/7/25/24205943/anthropic-ai-web-crawler-claudebot-ifixit-scraping-training-data |title=Anthropic’s crawler is ignoring websites’ anti-AI scraping policies|website=The Verge |language=en-US}}

{{reflist}}

• {{Official website}}

{{AI-based chatbots}}

{{Generative AI}}

{{Artificial intelligence navbox}}

Category:Machine learning

Category:Large language models

Category:Chatbots

Category:Virtual assistants

Category:2023 software