Anthropic launches Claude

Anthropic announced Claude, its first public-facing assistant, offering both a full model and a faster, cheaper “Claude Instant” variant through a chat interface and a developer API. The launch fell on the same day OpenAI released GPT-4, giving the field its first head-to-head release of two frontier chat models from separate labs, though Claude had already been running in closed access with launch partners including Notion, Quora and DuckDuckGo before the public announcement.

Anthropic pitched Claude on steerability and refusal behaviour rather than raw capability, describing it as trained to be “helpful, honest, and harmless,” language that traced directly to the Constitutional AI research the company had set out its rationale for days earlier — a technique for shaping model behaviour using a written set of principles and AI-generated feedback rather than relying solely on human raters. Early partner feedback, reported in Anthropic’s own announcement and picked up by outlets covering the launch, characterised Claude as less likely than competitors to produce harmful outputs and more conversational in tone, though these were not independently benchmarked comparisons at launch.

Claude’s arrival mattered less for besting GPT-4 on any measured capability — it did not attempt to — than for establishing Anthropic, founded by former OpenAI researchers specifically to pursue a different safety approach, as a genuine second pole in the frontier chatbot market rather than a research-only lab. That competitive framing hardened over the following months as both companies iterated rapidly; Anthropic’s next major release, Claude 2, arrived that July.