Claude Opus 5

Claude Opus 5 is Anthropic's July 2026 Opus release, and the pricing is the story: $5 per million input tokens and $25 per million output, identical to Claude Opus 4.8 and exactly half of Claude Fable 5 at $10/$50. The API id is claude-opus-5. It ships a 1 million token context window as both the default and the maximum, up to 128K output tokens, and text plus vision input. Anthropic's launch table puts it ahead of Fable 5 on most published rows despite the lower price: 70.6 percent on OSWorld 2.0 computer use against Fable 5's 66.1, 90.8 percent on BrowseComp against 87.4, 64.7 percent on Humanity's Last Exam with tools against 63.9, a GDPval-AA v2 knowledge-work score of 1861 against 1747, and 30.2 percent on ARC-AGI-3, a benchmark where Opus 4.8 scored 1.5. Fable 5 still edges it on FrontierCode (53.5 to 53.4) and DeepSWE (69.7 to 68.8), and Anthropic continues to position Fable 5 as the highest-capability tier. Treat every one of these as vendor-reported until independent runs land. Three operational changes matter for anyone porting from 4.8: adaptive thinking is now on by default when the thinking parameter is omitted, disabling thinking is only accepted at effort high or below (pairing it with xhigh or max returns a 400), and the minimum cacheable prompt drops to 512 tokens from 1024, so prompts previously too short to cache now create entries with no code change. Opus 5 draws on its own rate-limit bucket rather than the shared Opus 4.x pool, and Priority Tier does not cover it. Available on the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry.