Anthropic releases Claude Sonnet 5.5

Anthropic released Claude Sonnet 5.5, the second model in its Claude 5.5 line after Opus 5.5 six days earlier. The company described it as a faster, cheaper complement to Opus 5.5 rather than a new capability ceiling, aimed at “well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets.”

Pricing stayed at $2 per million input tokens and $10 per million output tokens — the same as Sonnet 5 — while Anthropic reported large gains on its own evaluations: 70.6% on the agentic-coding test Terminal-Bench 4.0, against Sonnet 5’s 10.3%; 46.2% on FrontierCode 1.1, against 42.4%; and 55.5% on CursorBench 4.0, against 34.1%. On GDPval-AA v2.1, Artificial Analysis’s knowledge-work index, Anthropic reported a score of 1,844 Elo, two points behind Opus 5.5 and far ahead of Sonnet 5’s 1,449. Anthropic also said Sonnet 5.5 was the first Sonnet model to beat the video game Pokémon Red working only from screenshots, and that because it typically needs fewer tokens to finish a task, it costs up to 30% less to run than Sonnet 5 despite the unchanged price.

Artificial Analysis, an independent evaluator, placed Sonnet 5.5 second on its own Intelligence Index at maximum reasoning effort — two points behind Opus 5.5 and 18 ahead of Sonnet 5 — but qualified Anthropic’s cost claim: at that top setting it measured the model using around 193,000 output tokens per task, the most of any model it had tested, putting its cost roughly 50% above Sonnet 5’s at the same setting even though the per-token price was identical. Anthropic’s own figures showed the cheaper picture holding at lower effort settings, where it said Sonnet 5.5 could match Sonnet 5’s best score for a fraction of the cost per task.

The model became available on Anthropic’s own platform and through Amazon, Google and Microsoft’s clouds as claude-sonnet-5-5. Anthropic said a smaller Haiku 5.5 model would join the line “in the coming weeks.”