Anthropic releases Claude Opus 5, its most agentic and best coding model yet

7:15 AM · July 27, 2026

Anthropic released Claude Opus 5 on July 24, 2026, priced at $5 per million input tokens and $25 per million output tokens, the same rate as the prior Opus model despite the capability jump. On FrontierBench v0.1, an evaluation built around turning engineering drawings into working software, Opus 5 scored 43.3 percent at maximum effort, more than double its predecessor's result and ahead of every other model Anthropic tested, including Moonshot's Kimi K3 and OpenAI's GPT-5.6 Sol. Anthropic pitched the release around frontier-class agentic coding and computer use, aimed at long-running coding agents and professional knowledge work rather than a single benchmark trick. The company also said Opus 5 recorded the lowest rate of deceptive behavior of any model in its automated safety audits to date, a metric it has increasingly foregrounded amid outside scrutiny of frontier labs' safety claims. Opus 5 becomes the default model on Claude Max and the top-tier option on Claude Pro, arriving less than two weeks after rival OpenAI's GPT-5.6 and just before Moonshot AI's Kimi K3 open-weight release, intensifying an already crowded July for frontier model launches.