Anthropic
Claude Opus 5.5 matches Fable 5.1 performance at 40% lower cost
Anthropic released Claude Opus 5.5 in general availability Sept. 22, 2026, matching Fable 5.1 on most work while cutting the cost to run it 40% versus Opus 5.
Anthropic introduced Claude Opus 5.5, calling it the first model in a new “Claude 5.5” family and its new leading model, and says it performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5.
The company made the announcement Sept. 22, 2026. Opus 5.5 is a general-availability release, not a preview or limited-access rollout, Anthropic said, and is now available on all platforms, including Amazon Web Services, Google Cloud and Microsoft Azure. Developers can call it on the Claude Platform using the API ID “claude-opus-5-5.”
Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, Anthropic said, a rate the company said is 20% below Opus 5. Cache reads cost $0.20 per million tokens, which Anthropic said is 60% less than Opus 5. A faster “fast mode,” available in Claude Code and the Claude Platform, runs up to 2.5 times faster and costs $8 per million input tokens and $40 per million output tokens, according to the company.
The model has a 1 million token context window and a maximum output of 128,000 tokens, per Anthropic’s models overview documentation, checked Sept. 23, 2026.
Anthropic also said it is increasing five-hour usage limits on its Pro, Max, Team and seat-based Enterprise subscription plans, and giving subscription users a rate-limit reset they can save and use later.
On benchmarks Anthropic reported alongside the launch, Opus 5.5 scored 1846 Elo at max effort on Artificial Analysis’s GDPval-AA v2.1 test, which evaluates agents on real-world professional work across 44 occupations, versus 1735 for Fable 5.1 and 1708 for Opus 5. On FrontierCode, which measures whether an agent’s code changes would be merged, Anthropic said Opus 5.5 scored 54.6% at its default, medium effort setting, above the 53.3% top score the company attributed to OpenAI’s GPT-6 Astra, at about one-fifth the cost per task.
Claude Code’s changelog lists Opus 5.5 as the CLI’s new default Opus model, in the entry for version 2.1.280, published to the npm registry Sept. 22, 2026.
Analysis
Every benchmark figure in Anthropic’s launch is Anthropic’s own number, run on a model Anthropic built. GDPval-AA v2.1 is Artificial Analysis’s test, but Anthropic chose which results to publish; FrontierCode’s comparison to GPT-6 Astra also comes from Anthropic’s post, not from OpenAI or an independent scorer. None of that makes the figures wrong, but where Opus 5.5 actually lands next to competing models on those same tests is not yet something an outside party has confirmed.
The more concrete story is the price. A 20% cut on standard tokens and a 60% cut on cached reads, paired with a claim of matching a newer model’s performance, reads as Anthropic trying to hold developers on Opus-tier pricing rather than pushing them toward a cheaper model down its own lineup. Defaulting Opus 5.5 into Claude Code within the same news cycle suggests the company wants that shift to happen immediately in its own tools, not on a timeline developers set themselves.
Opus 5.5 supersedes Opus 5 as Anthropic’s flagship Opus model with this release, but Anthropic has not said what happens to Opus 5 itself, and a lineup that adds a 5.5 tier without retiring anything is not a settled shape.