Anthropic Releases Claude Opus 5.5

Anthropic introduced Claude Opus 5.5, which matches Fable 5.1 performance while significantly reducing costs to $4 per million input tokens and $20 per output token.

Opus 5.5 is the first of the 5.5 generation, with Sonnet and Haiku to follow. Prices are 20% lower than Opus 5, and cache read costs have dropped by 60%.

Choi reports that Opus 5.5 outperformed competitors in most benchmarks, including Terminal-Bench 4.0 and GDPval-AA, though security guardrails were active during these evaluations.

Artificial Analysis data shows Opus 5.5 leading the Intelligence Index with 58 points, surpassing Fable 5.1 and GPT-6 Astra in several key metrics.

Anthropic highlights that the medium reasoning intensity setting offers superior performance-to-cost ratios, often outperforming higher intensity settings in specific benchmarks.

While Opus 5.5 max uses more tokens than its predecessor, its overall efficiency remains high. Anthropic has increased usage limits by 20% to reflect these gains.

In practical tests, Opus 5.5 proved faster and cheaper than Fable 5.1 for coding tasks, while showing better accuracy in report generation and bug detection.

Released just weeks after Fable 5.1, Opus 5.5 challenges the current market pricing. It is expected that competitors will need to adjust their pricing or performance soon.

Anthropic refined the model to be less verbose, prioritizing concise, actionable information. This improvement aims to reduce human review time and increase safety.

The release follows Dario Amodei’s call for AI safety. Opus 5.5 underwent rigorous external evaluation, showing improved behavior despite signs of the model recognizing it was being tested.

Choi notes that the industry is shifting toward cost-per-task efficiency. This trend makes it increasingly difficult to maintain high price points for premium AI models.

Full details regarding the Claude Opus 5.5 release are available on the official Anthropic website.