Updated on September 24, 2026
Claude Opus 5.5 is the first model in Anthropic's Claude 5.5 family, released on September 22, 2026. The company says it performs at Claude Fable 5.1 level on most work while running 40% cheaper than Opus 5. API rates start at $4 per million input tokens, and the model also ships inside the paid Claude plans. Its API identifier is claude-opus-5-5.
Adaptive thinking now runs on every request, with a max effort setting reserved for the hardest problems. Anthropic adds preserved thinking, a safeguard meant to make distillation by third parties harder, plus tougher resistance to prompt injection.
The other headline change is tone. Opus 5 had a reputation for long, jargon-heavy replies. Its successor puts the important information first and follows writing instructions more closely. You paste a messy Slack thread, it hands back three bullets your manager can actually read.
According to Anthropic, Opus 5.5 beats Fable 5.1 on every benchmark published at launch, from agentic coding to office-style knowledge work. The table below repeats the company's own figures, best read as vendor numbers until independent testing settles the debate.
At release it also took first place on the Artificial Analysis index, a handy reference if you follow LLM leaderboard rankings before picking a model.
| Benchmark | Opus 5.5 | Fable 5.1 | Opus 5 |
|---|---|---|---|
| Terminal-Bench 4.0 | 66.4% | 55.8% | 52.3% |
| FrontierCode v1.1 | 54.4% | 50.3% | 48.0% |
| CursorBench 4.0 | 57.8% | 51.8% | 46.6% |
| GDPval-AA v2.1 | 1,846 Elo | 1,735 Elo | 1,708 Elo |
| OSWorld 2.0 | 81.8% | 80.7% | 74.0% |
Input costs $4 per million tokens and output $20, down from $5 and $25 on Opus 5. Cache reads drop to $0.20, cache writes to $5, and fast mode runs at $8 and $40. These figures shift fast right now, so treat Anthropic's pricing page as the only source that counts.
The model is live in Claude, on the API platform, Amazon Bedrock, Google Cloud Vertex AI and Microsoft Azure, with Sonnet 5.5 and Haiku 5.5 slated for the coming weeks. Anthropic, founded in 2021 by former OpenAI researchers, ships this release right after Dario Amodei's public call to slow the capabilities race (the timing raised a few eyebrows).
That same day, OpenAI countered with GPT-6 Sol and Luna at roughly half the API price. The real fight now happens on cost per completed task, where Opus 5.5's token efficiency carries serious weight.
Claude Opus 5.5 is Anthropic's high-end model released on September 22, 2026, the first of the Claude 5.5 family. The company says it matches Claude Fable 5.1 on most work while costing 40% less to run than Opus 5. It targets coding, agents and professional knowledge work, under the API identifier claude-opus-5-5.
API pricing lands at $4 per million input tokens and $20 for output, 20% below Opus 5, with cache reads at $0.20. Inside the Claude apps, the model comes with the paid plans, whose five-hour session limits were raised alongside the launch.
On the nine benchmarks Anthropic published, Opus 5.5 scores higher than Fable 5.1 while charging $4 instead of $10 per million input tokens. Anthropic itself cautions that the real-world gap is narrower than the scores suggest, and Fable keeps an edge on some hard open-ended work. As a daily default, Opus 5.5 now makes the stronger case.
Both models launched the same day, about an hour apart. GPT-6 Sol undercuts Opus 5.5 by roughly half on API rates, while Anthropic's own tables put Opus 5.5 ahead on agentic coding, computer use and knowledge work, with Sol leading on workflow automation. Try Opus 5.5 first if long-running code is your daily bread.
Verdict: Top-shelf capability at a mid-shelf rate sums up the pitch. Teams running long-haul coding agents and Claude subscribers wanting the strongest house model gain on both fronts, speed and invoice alike.
GPT-6 Astra 
Claude Fable 5.1 
Claude Fable 5 
