Claude Opus 5.5 cuts list prices 20% and claims Fable 5.1 performance on most work

Anthropic released Claude Opus 5.5 on 22 September 2026. It is the first model in the Claude 5.5 family, and Anthropic says it performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5. API pricing drops to $4 per million input tokens and $20 per million output tokens, down from $5 and $25.

Sonnet 5.5 and Haiku 5.5 were described as coming “in the coming weeks.” The model is available through the Claude API, Claude Code, the Claude apps, and Amazon Web Services, Google Cloud and Microsoft Azure, under the model ID claude-opus-5-5, according to TechNews.

Two kinds of cheaper

Anthropic makes two separate cost claims, and they are easy to blur together.

The first is list price. Input and output tokens are each 20% lower than Opus 5. Cache reads fall from $0.50 to $0.20 per million tokens, and cache writes from $6.25 to $5. That is a straightforward price cut.

The second is the 40% figure, which Anthropic describes as the cost of typical workloads. Anthropic also says Opus 5.5 is the first model it would default to at medium effort. The post does not break down how much of the 40% comes from lower prices and how much from the model using fewer tokens per task, so it is safest to treat it as a figure for Anthropic’s own typical workloads, not a guarantee for yours.

Anthropic also says output is more than 30% faster than Opus 5. A fast mode, available in Claude Code and the Claude Platform, offers up to 2.5 times the speed at $8 input and $40 output per million tokens.

What the benchmarks say, and what Anthropic says about them

Anthropic published nine results comparing Opus 5.5 with Fable 5.1 and Opus 5. The company reports Opus 5.5 ahead on all of them:

• Terminal-Bench 4.0: 66.4%, against 55.8% for Fable 5.1 and 52.3% for Opus 5.
• FrontierCode v1.1 (Main): 54.4%, against 50.3% and 48.0%.
• OSWorld 2.1 (partial): 81.8%, against 80.7% and 74.0%.
• Humanity’s Last Exam with tools: 67.7%, against 65.6% and 63.6%.
• Terminal-Bench-Science 0.1: 58.7%, against 52.6% and 29.0%.

The biggest jump over Opus 5 is the science terminal benchmark, which nearly doubles. The smallest margins are on OSWorld and Humanity’s Last Exam, where Opus 5.5 leads Fable 5.1 by about one to two points.

These are vendor-run numbers, and the post carries its own caution. In Anthropic’s words:

“In our own use, the gap between Opus 5.5 and Claude Fable 5.1 is narrower than these scores suggest.”

That is an unusual sentence to put next to a benchmark table, and it is worth taking at face value. A reasonable reading is that Opus 5.5 is a cheaper way to get most of Fable 5.1’s results, not a replacement for it on every hard task. The TechNews write-up reports the announcement but gives no independent test results, and none had been published when it went up.

Changes that can break existing code

Two details in the release notes matter to developers more than the headline.

Thinking mode can no longer be disabled on Opus 5.5. If your application relied on switching reasoning off to save latency or tokens, you will need to control cost through the effort setting instead. Preserved thinking applies to API accounts created on or after 31 August 2026, so older accounts may not see the same behaviour.

Safeguards also differ from what some teams expect. Opus 5.5 ships with the same cybersecurity and biology protections as Fable 5.1. Most cybersecurity tasks are routed to Opus 4.8, and Anthropic says its Cyber Verification Program will extend to Opus 5.5 in the coming weeks. Biology research access requires the Life Sciences Verification Program. Security teams that depend on the model for defensive work should check that they are enrolled before they plan a migration.

Who should move

For people paying per token, the case is simple: same or better results than Opus 5 at a lower list price, with the caveat that you should re-run your own evaluations rather than trust the launch table. The 40% saving is the number to test, because it is the one that depends on your workload.

For subscribers, Anthropic raised five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans, and gave subscription users a rate limit reset they can save for later. The company did not say how large the increase is in the post.

Teams on Fable 5.1 have the harder decision. If Anthropic’s own remark holds, Opus 5.5 should cover most of their work for less money, but the hardest tasks are exactly where the gap is least understood. A sensible plan is to route routine jobs to Opus 5.5 and keep a sample of difficult ones on Fable 5.1 until you have your own comparison.

Still unknown: independent benchmark results, and how much of the 40% saving survives outside Anthropic’s own workloads.