Claude is the first publicly released model family from Anthropic, announced on March 14, 2023 in two versions: Claude, described by the company as "a state-of-the-art high-performance model," and Claude Instant, "a lighter, less expensive, and much faster option" (Introducing Claude (Anthropic, March 14, 2023)). It followed a closed alpha of several months with Notion, Quora and DuckDuckGo, and was made available through a chat interface and an API in Anthropic's developer console.
| Field | Value | |
|---|---|---|
| Developer | [[companies/anthropic | Anthropic]] |
| Announced | March 14, 2023 | |
| Versions | Claude; Claude Instant | |
| Access | Chat interface and developer-console API | |
| Parameters | Not disclosed | |
| Open weights | No | |
| Published benchmarks | None in the launch announcement |
Capabilities as stated at launch
Anthropic described Claude as "a next-generation AI assistant based on Anthropic's research into training helpful, honest, and harmless AI systems," capable of "a wide variety of conversational and text processing tasks while maintaining a high degree of reliability and predictability." The named use cases were summarization, search, creative and collaborative writing, question answering and coding (Introducing Claude (Anthropic, March 14, 2023)).
The launch claims were comparative and customer-reported rather than measured: Anthropic stated that early customers reported Claude "is much less likely to produce harmful outputs, easier to converse with, and more steerable," and that the model "can also take direction on personality, tone, and behavior." No benchmark scores, parameter counts, training-compute figures or evaluation results were published with the announcement, and no system card accompanied it — a contrast with the documentation practice Anthropic adopted for later generations such as Claude Opus 4.7.
Relation to Constitutional AI
The "helpful, honest, and harmless" formulation ties the product to Anthropic's published training research; the technique is set out in Constitutional AI: Harmlessness from AI Feedback and described at Constitutional AI. Claude 1 is the first deployed artifact of that programme, and the launch announcement is the point at which the research framing became a product claim.
Initial deployments
The launch announcement recorded six named partners, which together form the earliest record of commercial Claude deployment: Quora, serving Claude through its Poe app; Notion, for Notion AI's writing and summarization; DuckDuckGo, for the DuckAssist instant-answer feature drawing on Wikipedia and related sources; Juni Learning, for a Discord tutoring bot; Robin AI, for contract review and alternative-language suggestion (see Legal Services — AI Deployment); and AssemblyAI, for audio transcription and understanding APIs (Introducing Claude (Anthropic, March 14, 2023)).
Open questions
- Anthropic did not disclose parameter counts, training compute, context length or pricing in the launch announcement, and no such figures for Claude 1 are recorded here.
- The relationship between Claude 1 and Claude 2, announced July 11, 2023, is not yet documented; the Claude 2 announcement has not been ingested.
Relationships
- instance-of: Constitutional AI — the first model shipped under that training approach.
- related: Anthropic — the developer.
- related: Introducing Claude (Anthropic, March 14, 2023) — the launch announcement.
- depends-on: Constitutional AI: Harmlessness from AI Feedback — the research the product implements.
- related: Claude Opus 4.7 — a later generation of the same family, with published evaluations the first release lacked.
- related: Bard (Google) — a competing assistant launched in the same period.