Anthropic released Claude Opus 5.5 on 2026-09-22, three weeks after Claude Fable 5.1 and Mythos 5.1. The company’s headline claim is that Opus 5.5 “performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5,” and that it generates output more than 30 percent faster than Opus 5. It is available on the Claude Platform as claude-opus-5-5 and on AWS, Google Cloud and Microsoft Azure, with adjustable effort levels (low, medium as the default, high and max).
The published comparison against Opus 5 shows gains across agentic benchmarks: Terminal-Bench 4.0 at 66.4 percent against 52.3, FrontierCode v1.1 at 54.4 against 48.0, CursorBench 4.0 at 57.8 against 46.6, GDPval-AA v2.1 at 1846 Elo against 1708, and OSWorld 2.0 (partial reward) at 81.8 against 74.0. List prices dropped from Opus 5: 4 dollars per million input tokens (from 5), 20 dollars per million output (from 25), cache reads at 0.20 dollars (from 0.50) and cache writes at 5 dollars (from 6.25). Anthropic also cites one early customer completing a 680,000-line code migration in under a day.
The release carries the gating now standard at Anthropic. For users without verification, “most cybersecurity tasks will be re-routed to Opus 4.8,” with verified defenders admitted through an expanded Cyber Verification Program; biology research use sits behind a Life Sciences Verification Program, as with Fable 5.1. There are also behavioural changes: thinking can no longer be switched off, and preserved thinking is enabled for API accounts created after 2026-08-31, which Anthropic ties to distillation protection.
Why it matters: this is the clearest instance yet of a mid-tier model absorbing the previous flagship tier’s capability at a lower price within weeks, and it landed the same day as OpenAI’s GPT-6 Sol and Luna, which made the same cost-per-capability pitch. What it does not show: every benchmark number is Anthropic’s own, the “40% less to run” figure is a claim about typical workloads rather than a published measurement protocol, and “Fable 5.1 level on most work” leaves unstated where it falls short.