TL;DRClaude Opus 5 tops Fable 5 on most benchmarks at half the token price and is Anthropic’s most aligned model to date
Anthropic on Thursday released Claude Opus 5, a model that matches or exceeds its flagship Fable 5 on most coding and knowledge-work benchmarks at half the token price. The model is the fourth Anthropic has shipped in under two months, following Fable 5 in early June, Sonnet 5 at the end of June, and now Opus 5 on July 24. It costs $5 per million input tokens and $25 per million output tokens, the same price as its predecessor.
On Frontier-Bench, a coding evaluation that measures whether models can build working software from engineering drawings, Opus 5 more than doubles its predecessor’s score and surpasses every competing model, including Fable 5. On ARC-AGI 3, which tests novel problem-solving, Opus 5 scores roughly three times higher than the next-best model. On Zapier AutomationBench, which measures whether models can complete real business tasks end to end, Opus 5 passes more tasks than any rival even at its lowest effort setting.
Anthropic says Opus 5 is also its most aligned model to date, with the lowest rate of deceptive behaviour recorded in its automated safety audit. Its predecessor set a new bar for honesty when it launched in late May, and Opus 5 pushes that bar further while remaining behind Mythos 5 on offensive cybersecurity tasks. Anthropic says it intentionally avoided training the model on cyber work, though it has improved on those tasks as a side effect of becoming more generally capable.
