Anthropic released Claude Opus 5.5 on 22 September 2026 as the first model in its Claude 5.5 family, and the headline number is unusual: it is both more capable and cheaper than the model it replaces. On Anthropic’s published figures it costs 40% less to run than Opus 5 on typical workloads.
If you only remember one thing, make it this. The company says Opus 5.5 performs at the level of Claude Fable 5.1 on most work, which was previously the top of the range, while pricing sits well below Opus 5.
What you actually pay
Anthropic’s published pricing for Opus 5.5 is $4 per million input tokens and $20 per million output tokens. That is 20% below Opus 5 on both.
The number that matters most for agentic and coding work is cache reads, which Anthropic says make up the majority of costs in those workloads. Those are $0.20 per million tokens, a 60% reduction from Opus 5.
Output also generates more than 30% faster than Opus 5, so the same task finishes sooner and bills less.
The benchmark gap is large
On Terminal-Bench 4.0, an agentic coding benchmark, Opus 5.5 scores 66.4%. For context, Anthropic’s own comparison table lists Claude Fable 5.1 at 55.8%, Opus 5 at 52.3%, GPT-6 Astra at 57.9% and GPT-5.6 Sol at 37.3%.
FrontierCode v1.1 comes in at 54.4%, and CursorBench 4.0 at 57.8%. On GDPval-AA v2.1, a knowledge-work measure, Opus 5.5 scores 1846 against 1735 for Fable 5.1 and 1708 for Opus 5.
Anthropic is careful to add a caveat worth repeating: at these capability levels, benchmark margins have become a less reliable guide to real-world differences, and the company says the practical gap between Opus 5.5 and Fable 5.1 is narrower than the scores suggest.
Three things testers reported
Anthropic describes several early-tester results. One completed a 680,000-line code migration in under a day, work the company says would have taken an engineering team weeks. When asked to cut load times across every page of a web app, Opus 5.5 succeeded in 39 of 40 attempts, where Opus 5 made smaller improvements that also changed how the app behaved. A third tester had multiple Claude models build a game from a single prompt, and Opus 5.5 scored highest on graphics and polish.
Treat those as vendor-reported results rather than independent verification. Anthropic does say the model was tested before release by external evaluators including Frontier Design and METR.
The safety and compliance changes you need to know about
Three details matter if you run this in production.
Thinking cannot be switched off. Opus 5.5 is no longer available with thinking mode disabled, which changes token accounting for anyone who tuned costs by turning it off.
Preserved thinking is on. This is Anthropic’s anti-distillation safeguard, introduced with Fable 5.1. It prevents API users from editing Claude’s prior context to extract its reasoning, and it applies to API accounts created on or after 31 August 2026.
Watermarking applies. As with Fable 5.1, Opus 5.5 ships with Anthropic’s text watermarking measures to comply with the EU AI Act. If you publish model output verbatim, read the watermark documentation before you assume it is undetectable.
Who should switch, and who should wait
Switch now if you run agentic coding, long-context knowledge work, or anything with heavy cache reads. The combination of a 60% cut on cache reads and faster generation is the clearest win, and it is measurable in a week.
Test before committing if your prompts were tuned with thinking disabled. That configuration no longer exists, so your cost model will move even though the per-token price dropped.
Wait if you are mid-migration on Opus 5 and cannot absorb a behaviour change. Anthropic’s own guidance is that the model is stronger at finding and fixing inefficiencies, which means it will sometimes change things you did not ask it to change. The 39-of-40 load-time result cuts both ways.
Where to get it
Opus 5.5 is available on all platforms including Amazon Web Services, Google Cloud and Microsoft Azure. On the Claude Platform the model identifier is claude-opus-5-5, and Anthropic publishes a migration guide.
Anthropic says Claude Sonnet 5.5 and Claude Haiku 5.5 follow in the coming weeks with many of the same improvements to performance, efficiency and safety. If you are on a smaller budget, that is the release to watch, because Sonnet and Haiku are where most production workloads actually live.
Sources
Pricing and benchmark figures are as published by Anthropic on 22 September 2026. Verify current pricing on the official site before making purchasing decisions.
