Claude Opus 5

Anthropic's new flagship, under two months after Opus 4.8 and a larger jump than the recent point releases. AA Intelligence Index v4.1: 61 (#1 of 190, up from Opus 4.8's 56). The headline is coding and agentic work: Anthropic reports Opus 5 more than doubles Opus 4.8 on Frontier-Bench v0.1, lands within 0.5% of Fable 5's peak on CursorBench 3.2 at half the cost, surpasses Fable 5 on OSWorld 2.0 at roughly a third of the cost, and hits ~1.5× the next-best model on Zapier's AutomationBench. On reasoning it scores about 3× the next-best model on ARC-AGI 3, and on science it gains +10.2pp (organic chemistry) and +7.7pp (protein prediction) over Opus 4.8. API name claude-opus-5.

Priced at $5 / $25 per 1M input/output tokens with a 1M-token context (per Artificial Analysis). Carries forward the effort control and ~2.5× fast mode (at 2× base price), and adds beta mid-conversation tool changes and automatic API fallbacks. Available on claude.ai, Claude Code, Claude Cowork, and the Claude API.