Anthropic has launched Claude Opus 5, and the company frames it as a leap rather than a routine update. It calls the model a step-change improvement over Claude Opus 4.8, with the biggest gains in deep reasoning, agentic coding, and long-horizon tasks. For developers and businesses leaning on AI for complex work, the headline is capability at a notably lower price.
Anthropic says Opus 5 delivers frontier intelligence at half the cost of Claude Fable 5, its most capable widely available model. The rate itself holds steady at $5 per million input tokens and $25 per million output, unchanged from Opus 4.8. So users get a meaningfully stronger model without paying more, which is unusual in a market used to premium pricing for each new tier.
On Opus 4.8, requests ran without extended reasoning unless developers switched it on. Opus 5 flips that, letting the model decide when and how deeply to think on each turn. That should improve answers out of the box, though it also means developers must revisit their output-token limits, since thinking and response text now share the same budget.
Opus 5 converts extra effort into better results more reliably than any earlier Opus model, across a ladder running from low up to a new “max” tier for the deepest reasoning. There is a catch worth noting, since disabling thinking now works only at high effort or below. Asking for no thinking at the top tiers simply returns an error, a deliberate design choice.
Anthropic highlights stronger agentic coding that completes multi-file features without leaving stubs, sharper code review that finds real bugs with few false positives, and better vision for charts, documents, and UI work. The model also handles a 1M-token context window as standard, alongside improved spreadsheet and slide generation.
Some quieter behavior changes matter for anyone upgrading. Opus 5 writes longer responses by default, narrates its progress more during agentic sessions, and verifies its own work without being asked. Anthropic even advises removing old “add a verification step” instructions, since they now cause the model to over-check. It is available immediately on the Claude API, AWS, Google Cloud, and Microsoft Foundry.
Claude Opus 5 is priced at $5 per million input tokens and $25 per million output tokens, unchanged from Claude Opus 4.8.
