Anthropic Claude 2
Anthropic positioned Claude 2 as a major step up from its first-generation assistant, headlined by a 100,000-token context window — roughly 75,000 words, or hundreds of pages — that let users feed in whole books or large document sets. It claimed sharply improved coding (71.2% on the HumanEval Python test, up from 56.0%) and better math and reasoning, plus the ability to draft longer documents in a single pass. The pitch leaned on Anthropic's safety framing, with Claude trained to be more 'helpful, harmless, and honest' than rivals.
Claude 2 launched broadly available the same day via a new public beta chat website (claude.ai) for users in the US and UK, alongside API access for businesses, with the 100K context window live at launch.
Claude 2 was Anthropic's real public debut — its first product anyone could try without a waitlist — and the 100K context window was a genuine differentiator at a time when most rivals were capped far lower. In practice it was a credible alternative to ChatGPT rather than a leader: capable on long documents and writing, but generally seen as a notch below GPT-4 on hard reasoning, and the cautious safety tuning drew complaints about over-refusal. Its strategic value was establishing Anthropic as a serious independent lab and a viable second supplier for enterprises wary of single-vendor dependence on OpenAI. The long-context bet later proved directionally right as the whole industry chased larger windows.
It marked Anthropic's transition from research lab to a real product company and gave enterprises a credible second source to OpenAI — the start of a genuine two-horse race at the frontier.