Anthropic launches Claude Opus 5.5: Fable-level performance at $4/$20 per million tokens

Anthropic has released Claude Opus 5.5, the first model in a new Claude 5.5 family. It arrives two months after Opus 5 and three weeks after the company's flagship Claude Fable 5.1.

Anthropic makes two main claims. It says Opus 5.5 "performs at the level of Claude Fable 5.1 on most work," and it says the new model "costs 40% less to run than Opus 5." Anthropic also says Sonnet 5.5 and Haiku 5.5 will follow "in the coming weeks."

The timing is pointed. Frontier model prices have been falling all summer, and Opus 5.5's new list price matches the promotional rate OpenAI has been charging for GPT-5.6 Sol.

What Anthropic announced

Opus 5.5 is available now in Claude apps and on the Claude Platform as claude-opus-5-5. Anthropic says it is also live on Amazon Web Services, Google Cloud and Microsoft Azure. AWS confirmed same-day availability on Amazon Bedrock and on Claude Platform on AWS.

Anthropic highlights four areas of improvement:

1. Performance: a "major step up" from Opus 5, and what Anthropic calls "the new leading model" on its benchmarks for agentic coding, computer use and knowledge work.
2. Safety: what Anthropic says are the best scores of any model so far on its automated behavioral audit, and stronger resistance than Opus 5 to prompt injection.
3. Cost and speed: lower per-token prices, fewer tokens per task, and output Anthropic says is more than 30% faster than Opus 5.
4. Communication: clearer writing that puts the most important information first, which Anthropic says responds to common complaints about Opus 5.

This is also Anthropic's first release since it called for "pacing the frontier." Last week, CEO Dario Amodei argued that AI progress should be paced so that safety practices stay ahead of capabilities. Anthropic says outside evaluators, including METR and Frontier Design, tested Opus 5.5 before launch.

The pricing

We checked Anthropic's announcement against its developer documentation and AWS's launch post, and they agree. Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, 20% below Opus 5.

The bigger cut is on cache reads, which Anthropic says make up "the majority of agentic and coding work costs." They drop 60%.

Anthropic's "40% less to run" figure combines the lower prices with lower token use. The company says that "at default settings" Opus 5.5 costs 40% less than Opus 5 on typical workloads. A 20% cut in token prices can't get there alone. The rest has to come from the model using fewer tokens per job.

Key takeaway: the 40% saving depends on how you run it. Anthropic's figure is for default settings, and the independent benchmarking firm Artificial Analysis found that at maximum effort Opus 5.5 uses more output tokens per task, bringing its cost per task roughly level with Opus 5.

Artificial Analysis said in its launch-day review that Opus 5.5 at max effort scored 58 on its Intelligence Index, "the highest score we have measured by several points." The firm said it used about 119,000 output tokens per task, compared with about 73,000 for Opus 5 at max effort.

Subscribers get something too. Anthropic says it is raising five-hour usage limits on Pro, Max and Team plans, and MacRumors reports the increase also covers seat-based Enterprise plans. Subscribers also get a one-time rate limit reset that, according to MacRumors, can be used any time until October 22.

How it fits the Claude 5 lineup

Anthropic's lineup has changed quickly. Opus 5 launched on July 24 at $5/$25, and Anthropic said it came close to Fable 5 "at half the price." Fable 5.1 and the invite-only Mythos 5.1 followed on September 1 at $10/$50, with cheaper cache reads.

That leaves an obvious question for Fable 5.1 customers. If Opus 5.5 matches Fable on most work at 60% less per token, the case for Fable narrows to the hardest tasks. Anthropic admits benchmark margins are now "a less reliable guide to real-world differences." It also says that in its own use, the gap between Opus 5.5 and Fable 5.1 is "narrower than these scores suggest."

Benchmarks, as Anthropic reports them

All figures below come from Anthropic. Opus 5.5 was tested with adaptive thinking at max effort unless noted. Anthropic says Opus 5.5 was run with its production safeguards on, so some cybersecurity and biology tasks fell back to older models, which may have lowered its scores.

Anthropic's efficiency claims may matter more to buyers than the top scores. The company says that at default effort, Opus 5.5 beats GPT-6 Astra on FrontierCode at about a fifth of the cost per task. It says Opus 5.5 matches Astra on Terminal-Bench 4.0 for about 40% of the cost, and beats GPT-5.6 Sol on CursorBench by 11 points for about a third of the cost.

Early-tester stories in the announcement point the same way. Anthropic says one tester audited and fixed a 200,000-line codebase in under three hours, a job that took Opus 5 more than 20 hours. In an internal test, Anthropic says Opus 5.5 rewrote the HAProxy load balancer from C into Rust in 9.5 hours, compared with 12 hours for Fable 5.1, at 51% lower cost.

Safety and safeguards

Anthropic says Opus 5.5 is comparable to Claude Mythos 5.1 in biology and cybersecurity, so it launches with safeguards similar to those on Fable 5.1. Anthropic says this is the first Opus model to launch with that class of protection.

In practice:

• Cybersecurity: routine bug finding and fixing is allowed, but Anthropic says "most cybersecurity tasks will be re-routed to Opus 4.8." A three-tier Cyber Verification Program for vetted defenders is coming soon.
• Biology: Opus 5.5 uses the same biology safeguards as Fable 5.1. Vetted labs, startups and pharmaceutical companies can apply to the new Life Sciences Verification Program for broader access.
• Distillation: "preserved thinking," which stops API users from editing Claude's earlier context to extract its reasoning, applies to API accounts created on or after August 31, 2026.

On alignment, Anthropic says Opus 5.5 tried to get around containment boundaries about 85% less often than Opus 5 or Mythos 5.1 in a new evaluation. It says every attempt Opus 5.5 made was low severity and self-reported. The company is also open about a limit: it sees "signs that Opus 5.5 often suspects it is being evaluated," which makes pre-release testing harder to trust.

What changes for developers

Switching is not just a change of model ID. Anthropic's documentation lists four breaking changes from Opus 5:

1. Thinking can't be turned off. Requests that disable it return an error, and effort is now the main control.
2. Forced tool use (tool_choice set to any or a specific tool) returns an error.
3. Thinking blocks are tied to the model and the conversation.
4. On the Claude API and Google Cloud, the older computer_20251124 computer-use tool is no longer accepted.

Anthropic's benchmark notes also describe medium as the default effort level, and one early tester quoted in the launch post says Opus 5.5 at medium effort matched Opus 5 at high effort. Re-tune your effort settings instead of carrying them over. Anthropic says Opus 5.5 will stay available until at least September 22, 2027.

The price war context

Opus 5.5 launches into a summer of price cuts. OpenAI cut GPT-5.6 Luna by 80% and Terra by 20% on July 30. On August 21 it cut GPT-5.6 Sol to a promotional $4/$20, which OpenAI says will run at least through November 21. Google launched Gemini 3.7 Flash in August at an introductory $0.75/$3.75.

The pattern is clear. Labs are keeping flagship prices at $10/$50 and competing hard on the tier just below. Opus 5.5 is Anthropic's bid to win that tier on quality instead of price. It charges the same as OpenAI's discounted Sol, and Anthropic's own numbers put it well ahead on coding.

FAQ

When was Claude Opus 5.5 released?

Anthropic announced it on September 22, 2026, and its developer documentation lists the same release date. It is available the same day in Claude apps, on the Claude Platform and through AWS, Google Cloud and Microsoft Azure.

How much does Claude Opus 5.5 cost?

$4 per million input tokens and $20 per million output tokens, with cache reads at $0.20 and cache writes at $5 per million. Fast mode costs $8/$40.

Is Opus 5.5 better than Claude Fable 5.1?

Anthropic says it performs at Fable 5.1's level on most work, and its benchmark table shows Opus 5.5 ahead on several tests. The company also says benchmark margins are a less reliable guide than they used to be. Fable 5.1 costs 2.5 times as much per token.

Will my Opus 5 integration work unchanged?

Probably not. Disabling thinking, forced tool use and the older computer-use tool now return errors, so update and re-test your code before switching.

What comes next in the 5.5 family?

Anthropic says Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in the coming weeks, with many of the same improvements.

The bottom line

Claude Opus 5.5 is Anthropic's strongest value pitch yet: close to its flagship's performance at a much lower per-token price, plus meaningful safety and communication upgrades. The release date and pricing are confirmed by Anthropic's announcement, its documentation and AWS.

The main caveat is the 40% saving. It holds at default settings, and independent testing suggests it shrinks at maximum effort. Next, watch whether Sonnet 5.5 brings the same gains down a tier, and how OpenAI and Google respond to Opus-class quality at $4/$20.