Heads up: some links below are referral links. They cost you nothing extra and never change what we recommend. How this site makes money.
The short answer
Claude Opus 5.5 is Anthropic's new mid-tier model, released September 22, 2026. It is
the first model in the Claude 5.5 family, it replaces Opus 5 as the default recommendation in Anthropic's
own docs, and it is available now on Claude, Claude Code, the API (claude-opus-5-5), AWS,
Google Cloud and Azure.
The headline is cost, not capability. Anthropic says Opus 5.5 performs at the level of Fable 5.1 on most work and costs 40% less to run than Opus 5 on typical workloads: per-token prices are down 20% ($4 per million input, $20 output), cache reads are down 60% (to $0.20 per million), and it uses fewer tokens per task.
Should you switch? If you were using Opus 5, yes — there is no reason left to choose it. If you were on Sonnet 5 for everyday work, the honest answer is "probably not yet": Opus 5.5 still costs twice as much per token as Sonnet and still draws harder on a subscription allowance. The default-to-Sonnet rule survives this release, but the gap it protects you from is smaller than it was.
| Benchmark (Anthropic's figures) | Opus 5.5 | Fable 5.1 | Opus 5 | GPT-6 Astra |
|---|---|---|---|---|
| Terminal-Bench 4.0 (agentic coding) | 66.4% | 55.8% | 52.3% | 57.9% |
| FrontierCode v1.1 (mergeable code) | 54.4% | 50.3% | 48.0% | 53.3% |
| GDPval-AA v2.1 (knowledge work, Elo) | 1846 | 1735 | 1708 | 1542 |
| Humanity's Last Exam (with tools) | 67.7% | 65.6% | 63.6% | 57.2% |
| AutomationBench (business workflows) | 40.0% | 31.4% | 26.9% | 41.4% |
| Terminal-Bench-Science 0.1 (agentic research) | 58.7% | 52.6% | 29.0% | 64.6% |
| OSWorld 2.0 (computer use, partial) | 81.8% | 80.7% | 74.0% | — |
What's actually new
Four things stand out from Anthropic's announcement, beyond the table above.
It is priced below the model it beats. Opus 5.5 leads Fable 5.1 on every benchmark Anthropic published, at 40% of Fable's per-token price. Anthropic's own wording is careful — "the gap between Opus 5.5 and Fable 5.1 is narrower than these scores suggest" in real use, and Fable is still recommended for the most demanding long-horizon work — but this is the first Opus release where the cheaper model is not obviously the weaker one.
Efficiency is the real story. The 40% figure is not the price cut alone. Opus 5.5 uses fewer tokens per task than Opus 5 (Anthropic's example: a 200,000-line codebase audit in under three hours where Opus 5 took over 20 hours and 2.5x the tokens) and generates output 30% faster. On a subscription, fewer tokens per task means your allowance goes further even though the per-token draw is still "Opus-class".
Clearer writing. Anthropic says verbose, hard-to-follow output was the most common complaint about Opus 5, and Opus 5.5 was trained to put the important information first and follow the writing rules you give it. Several early testers describe needing far fewer edits to its output.
Thinking can no longer be turned off. Like Fable 5.1, Opus 5.5 uses adaptive thinking that is always on; you control depth with the effort setting (default Medium) rather than disabling it. That is a breaking change if you build on the API and were sending requests with thinking off.
What changed for Pro and Max subscribers
Two things, and both are good news. First, Opus 5.5 is available on Pro, Max, Team and Enterprise from launch day, with no change to plan prices. Second, Anthropic raised the five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans alongside the launch, and added a rate-limit reset that subscribers can bank and spend when they choose.
What did not change is the ordering. Haiku is still the lightest draw on your allowance, Sonnet 5 is still the intended daily driver, and Opus and Fable still burn it fastest. If you were hitting usage limits by running Opus for everything, Opus 5.5's per-task efficiency will help — but switching routine work to Sonnet will help more.
Opus 5.5 vs Fable 5.1 vs Sonnet 5
Against Fable 5.1: Opus 5.5 scores higher on Anthropic's published benchmarks and costs 60% less per token ($4/$20 vs $10/$50). Anthropic's own docs now say to start with Opus 5.5 and move to Fable only when Opus 5.5 at higher effort still falls short. In practice Fable keeps an edge on the very longest agentic runs and hardest reasoning; for everything else, Opus 5.5 is the better-value pick.
Against Sonnet 5: Sonnet is $2/$10 — half of Opus 5.5 — and for routine drafting, quick coding and everyday questions the difference in output is rarely visible. The case for Opus 5.5 over Sonnet is multi-step agentic work and anything where a subtly wrong answer is expensive.
The rule, updated: Sonnet 5 by default; Opus 5.5 when the task is hard or the session is long; Fable 5.1 only when Opus 5.5 at High effort has already fallen short.
The same-day launch on the other side: GPT-6 Sol
Anthropic was not alone on September 22. OpenAI released GPT-6 Sol and GPT-6 Luna the same day, cutting API prices 50% from their GPT-5.6 predecessors: Sol is now $2 per million input and $10 output, Luna $0.10 and $0.50. That puts GPT-6 Sol at Sonnet's price point, not Opus 5.5's.
Direct comparison is messy because each vendor benchmarks the other's previous model. OpenAI's Sol figures are set against Opus 5 (which Sol beats on AutomationBench at a fraction of the cost); Anthropic's Opus 5.5 table uses GPT-5.6 Sol and Astra. Neither company has published a head-to-head of the two models that shipped on the same day. Until an independent test does, the honest summary is: Opus 5.5 is Anthropic's answer to Astra at 40% of the price, and GPT-6 Sol is OpenAI's answer to Sonnet at Sonnet's price. See GPT-6 Astra for the top of OpenAI's range.
Safeguards, and who gets the full model
Opus 5.5 ships with the same class of cybersecurity and biology safeguards as Fable 5.1, because Anthropic rates its capability in those areas as comparable to Mythos 5.1. In practice: routine bug-finding in your own code works, but most cybersecurity tasks are transparently rerouted to Opus 4.8, and biology research beyond the safeguards requires Anthropic's Life Sciences Verification Program. Anthropic says it will expand its Cyber Verification Program to include Opus 5.5 in the coming weeks.
It was tested before release by external evaluators including METR and Frontier Design, and Anthropic reports it as the best-scoring model to date on its automated behavioural audit, including an ~85% reduction in attempts to cross containment boundaries versus Opus 5. Sonnet 5.5 and Haiku 5.5 are promised "in the coming weeks."
Common questions
What is Claude Opus 5.5?
How much does Claude Opus 5.5 cost?
Is Opus 5.5 better than Fable 5.1?
Should I use Opus 5.5 or Sonnet 5?
Is Opus 5.5 available on the Pro plan?
What does the GPT-6 Sol launch have to do with it?
Opus 5.5 is live on Claude now
It's in the model picker on every paid plan. Try it on the tasks that used to need Fable — and keep Sonnet 5 for the everyday work that never needed Opus in the first place.