Claude Opus 5.5 is Anthropic's newest flagship AI model for coding and knowledge work — matches their top-tier model on most tasks at 40% lower cost.
What Changed
Opus 5.5, released September 22, 2026, is the first model in Anthropic's new Claude 5.5 family. It performs at roughly the level of Anthropic's top model, Claude Fable 5.1, on most work, while costing 40% less to run and outputting responses over 30% faster than the previous Opus 5. It's a routine-sounding version bump on paper, but the release is framed differently from a typical model launch: it's Anthropic's first since CEO Dario Amodei publicly called for the industry to "pace" the frontier — slow capability gains down enough that safety work can keep up — and it shipped within hours of OpenAI's GPT-6 Sol and Luna. Anthropic backed that framing with external pre-release testing by two independent evaluators, Frontier Design and METR, and says Opus 5.5 attempted to bypass the boundaries it was given about 85% less often than Opus 5 or Claude Mythos 5.1 in a dedicated safety evaluation.
What It's Actually Good At
- An early tester used it to complete a 680,000-line code migration in under a day — work that would otherwise have taken an engineering team weeks.
- Asked to fix load-time inefficiencies across every page of a web app, it succeeded 39 out of 40 times; Opus 5 made smaller fixes that also changed the app's behavior.
- It audited and fixed a 200,000-line codebase in under 3 hours, versus over 20 hours and 2.5x the tokens for Opus 5.
- Rewriting HAProxy (a widely-used load-balancing tool) from C to Rust, it finished in 9.5 hours at 51% lower cost than Fable 5.1's 12-hour, more expensive rewrite — both passed nearly all of HAProxy's own regression tests.
- Asked to write a financial report using only hard-to-find sources, 16 of 18 of its reports passed a strict fact-check (no invented figures or quotes); neither Fable 5.1 nor Opus 5 passed on any attempt.
What Real Teams Are Saying
Anthropic published named quotes from engineers testing the model against real work, not anonymous praise:
- Cristian Rivera, Staff Software Engineer at Stripe — had one Opus 5.5 session direct a dozen more sessions through a multi-day rebase of 40 stacked pull requests; all 40 passed CI the next afternoon.
- Aabhas Sharma, CTO at Hebbia — reported it covered 86.6% of what their finance-workflow rubric looks for, versus 60.3% for Opus 5.
- Peter Albert, Co-Founder at Viktor — said their AI employee gets twice as many hard tasks right using nearly half the cost of Opus 5.
- Frank Corrao, Head of Central Equity Quant Research Engineering — noted it caught a one-off indexing error in their own evaluation instructions that no prior model had caught.
Pricing and Access
Input tokens cost $4 per million and output tokens $20 per million — 20% less than Opus 5. Repeated context (cache reads) costs $0.20 per million, a 60% cut. A faster mode running at up to 2.5x speed is available in Claude Code and the Claude Platform for $8/$40 per million tokens. The model is live now on the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry/Azure, with the model ID claude-opus-5-5; Claude Sonnet 5.5 and Haiku 5.5 are expected to follow in the coming weeks.