Claude 1
Last reviewed
Jun 3, 2026
Sources
11 citations
Review status
Source-backed
Revision
v1 · 1,406 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 3, 2026
Sources
11 citations
Review status
Source-backed
Revision
v1 · 1,406 words
Add missing citations, update stale details, or suggest a clearer explanation.
Claude 1 was the first generation of Anthropic's Claude family of large language models, launched in March 2023 and offered in two variants: a full-sized model called simply Claude and a faster, cheaper sibling called Claude Instant. [1][2] The line went through several incremental point releases (1.0 through 1.3) over its lifetime and introduced the 100,000 token context window that became one of the family's defining features. Unlike the consumer chatbots that arrived later, the first Claude was gated to approved companies and developers rather than the general public. [1][3] For a broader treatment of the model family across all of its generations, see Claude.
Anthropic was founded in January 2021 by seven former employees of OpenAI, among them the siblings Dario Amodei and Daniela Amodei. [4] Dario had been OpenAI's vice president of research and Daniela its vice president of safety and policy, and the group left in part over disagreements about direction and the priority given to safety. [4] The new company organized itself around research into AI safety and alignment, and spent roughly two years building and testing systems before putting a product in front of customers.
By the time Anthropic released Claude, the market had been reshaped by OpenAI's ChatGPT, which launched in late 2022 and made conversational language models a mainstream topic. Press coverage of Claude's debut consistently framed it as a rival to ChatGPT, and noted that Anthropic counted Google among its investors. [1][5]
Anthropic introduced Claude on March 14, 2023, opening early access to its developer API on the same day. [1][2] The company shipped two models. Claude was described as a high-performance model for demanding tasks, while Claude Instant was positioned as a lighter, less expensive, and much faster option for workloads that did not need the larger model. [2] Both were aimed at text work such as summarization, search, question answering, creative and collaborative writing, and coding. [2]
The release was deliberately narrow. TechCrunch reported that Anthropic was "not pursuing a broad direct to consumer approach at this time" and that organizations had to request access; pricing had not yet been detailed publicly at launch. [1] Anthropic had spent the preceding months running a closed alpha with a small group of companies, several of which it named as launch partners. [2]
| Partner | How it used Claude |
|---|---|
| Quora | Offered Claude to users through its Poe chat app |
| Notion | Built Claude into the Notion workspace as part of its AI assistant |
| DuckDuckGo | Powered DuckAssist, an Instant Answer feature that generated answers to search queries |
| Robin AI | Used Claude to review contract clauses and suggest alternative language |
| AssemblyAI | Used Claude in APIs that transcribe and understand audio at scale |
| Juni Learning | Powered its Discord-based Juni Tutor Bot for online tutoring |
Sources: Anthropic's announcement, with the partner roster also reported by TechCrunch. [1][2] Robin AI's chief executive, Richard Robinson, told reporters that among the technologies the firm had worked with since 2019, "nothing has matched Claude's capabilities." [2]
Claude 1 was not a single static model. Over 2023 Anthropic shipped a sequence of point releases under both the full and Instant lines, identified internally and in its API by version strings. The first generation eventually covered the model identifiers claude-1.0, claude-1.1, claude-1.2, and claude-1.3, along with claude-instant-1.0, claude-instant-1.1, and claude-instant-1.2. [6] These successive versions brought incremental improvements rather than new architectures, and Claude 1.3 became the high-water mark of the generation against which Anthropic later measured Claude 2. [7]
The most consequential change came on May 11, 2023, when Anthropic expanded Claude's context window from 9,000 tokens to 100,000 tokens, roughly 75,000 words. [3][8] The jump let users drop hundreds of pages of material, or an entire book, into a single prompt and ask questions that required synthesizing information across the whole text. [3] To demonstrate the capability, Anthropic loaded the full text of The Great Gatsby (about 72,000 tokens) into Claude Instant, changed a single line so that one character was described as a machine learning engineer at Anthropic, and the model identified the altered sentence in 22 seconds. [3] The new limit far exceeded what rivals offered general users at the time: ChatGPT then ran with about 4,000 tokens of context and even the largest GPT-4 variant topped out near 32,000. [9]
| Milestone | Date | Detail |
|---|---|---|
| Claude and Claude Instant announced | March 14, 2023 | API early access opens; partners only [1][2] |
| 9K token context window | At launch | Initial prompt size for the first models [3] |
| 100K token context window | May 11, 2023 | Expanded to roughly 75,000 words [3][8] |
| Claude 1.3 | 2023 | Final and strongest first-generation release [7] |
| Claude 2 supersedes the line | July 11, 2023 | Second generation; public claude.ai beta [7] |
Anthropic trained the first Claude with a combination of reinforcement learning from human feedback (RLHF) and its own method, Constitutional AI. [10][11] Rather than relying solely on human labelers to flag harmful outputs, Constitutional AI has the model critique and revise its own responses against a written set of principles, a "constitution." TechCrunch described the approach at launch as principle-based and grounded in a short list of values including beneficence, or maximizing positive impact; nonmaleficence, or avoiding harmful advice; and autonomy, or respecting freedom of choice. [1] Anthropic published the text of Claude's constitution in May 2023, drawing on sources that ranged from the UN Declaration of Human Rights to platform trust-and-safety guidelines. [11] The technique was one of the features Anthropic and the press repeatedly cited to distinguish Claude from competing systems. [5]
Coverage treated Claude as a serious entrant rather than a clear winner. Reviewers credited its conversational, easy-to-read answers and the safety framing of Constitutional AI, while noting limitations that Anthropic itself acknowledged. The company conceded that Claude could hallucinate, and TechCrunch reported it was weaker at math and a poorer programmer than ChatGPT, and that determined users could still prompt their way around its safeguards. [1] The 100,000 token window drew the most sustained attention after May 2023, because the ability to feed long documents, financial filings, or transcripts into a single prompt was something most competitors could not yet match for ordinary users. [3][9]
Anthropic released Claude 2 on July 11, 2023, about four months after the first generation. [7] TechCrunch characterized it less as a ground-up rebuild than as "a tweaked version of Claude 1.3," with reported gains in coding, math, and reasoning while keeping the 100,000 token context window introduced in the prior generation. [7] Claude 2 was also the first Claude model Anthropic offered directly to the public, through a free beta chat interface at claude.ai, ending the partners-only access model of the first generation. [7] The Claude 2 line was later followed by the third generation in March 2024, led by Claude 3 Opus.
The original models were eventually retired. Anthropic notified developers on September 4, 2024 that the Claude 1 and Claude Instant models were scheduled for retirement, and the claude-1.x and claude-instant-1.x identifiers were withdrawn from the API on November 6, 2024. [6] Their lasting contribution was less any single benchmark than the two ideas they carried into every later release: Constitutional AI as a training method, and the very long context window as a product feature.