A specimen sheet for Anthropic's most intelligent generally available model — a new tier above Opus, catalogued here in field notes, hard shadows, and five hidden things.
Extended reasoning that drafts the outline before writing the page. Long-horizon agentic work across tools, files, and hours — then an honest report of what actually happened.
OBS-001It writes code that reads like the surrounding code — your repo's idioms, your comment density, your naming. Less "AI wrote this," more "a careful colleague did."
OBS-002Vast context, carefully held. Page one still matters on page nine hundred.
OBS-003Fewer hallucinations, more "let me check." It asks when it must and decides when it can.
OBS-004Ships with additional safety measures for dual-use capabilities. The most capable Claude anyone can pick up — built so it's hard to pick up wrong.
OBS-005The generally available telling. Full Claude 5 intelligence with additional safety measures for dual-use capabilities — the story told in the village square.
The same underlying model, without those measures — available only to approved organizations. The unabridged manuscript in the locked wing of the library.