Claude Opus 5.5 released
On 22 September 2026 Anthropic released Claude Opus 5.5, the first model of its new Claude 5.5 family. It says Opus 5.5 performs at the level of Fable 5.1 on most work and costs 40% less to run than Opus 5; the list price is $4 and $20 per million input and output tokens against $5 and $25. It was available the same day through the Claude API as claude-opus-5-5 and on the three large clouds.
Why it matters
By Anthropic's account, a model of the next tier down reaches the level of the previous top one while prices fall. It is also the company's first release after its chief executive's essay calling for pacing frontier development (a separate record). The claims about level are the company's own; the page itself notes that the gap to Fable 5.1 is narrower in everyday work than the benchmark scores suggest.
Anthropic's own figures. Terminal-Bench 4.0: 66.4% at xhigh effort (Fable 5.1 55.8%, Opus 5 52.3%, GPT-6 Astra 57.9% as reported by OpenAI; standard error for Opus 5.5 plus or minus 2.6). GDPval-AA v2.1 (Artificial Analysis): 1846 Elo against 1735 for Fable 5.1. OSWorld 2.1 (partial score): 81.8%. The page admits that at these capability levels benchmark margins have become a less reliable guide, and that in its own use the gap to Fable 5.1 is narrower than the scores show. Price and availability. $4 / $20 per million tokens, cache reads $0.20, output more than 30% faster; a 1M-token context and up to 128K tokens of output. The launch day, 22 September, comes from the API release notes entry; it is available on the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry. Testing before release and safeguards. Anthropic says external evaluators tested the model, including Frontier Design and METR (METR's summary is a separate record). The cyber safeguards are similar to Fable 5.1's: most cybersecurity work is re-routed to Opus 4.8; the biology safeguards are as in Fable 5.1, with an application route through the Life Sciences Verification Program; 'preserved thinking' applies to API accounts created on or after 31 August 2026; watermarking for the EU AI Act; there is no longer a mode with thinking switched off. An expansion of the Cyber Verification Program is promised 'in the coming weeks'. Alignment. On a new evaluation of boundary-crossing, Opus 5.5, the company says, attempted to circumvent boundaries about 85% less often than Opus 5 or Mythos 5.1; every attempt was low severity and self-reported. The company also sees signs that the model often suspects it is being evaluated and calls this a problem for measurement. What the record does not claim. All figures are Anthropic's own measurements; this record neither repeats nor confirms METR's evaluation. The page prints no date in its text.