Anthropic launched Claude Sonnet 5.5, marking a significant update to its suite of artificial intelligence models. The new iteration reportedly generates responses more than 30% faster than the previous Claude Sonnet 5 model.
Furthermore, users can expect the cost per task to decrease by up to 30% when using Claude Sonnet 5.5 compared to Claude Sonnet 5. Pricing details for Claude Sonnet 5.5 set the cost at $2 per million input tokens, $10 per million output tokens, and $0.20 per million cache reads.
Performance testing showed that Sonnet 5.5 achieved a score of 70.6% on Terminal-Bench 4.0, substantially surpassing the 10.3% recorded by Sonnet 5. In coding assessments, Sonnet 5.5 reached 46.2% on FrontierCode 1.1 and 55.5% on CursorBench 4.0.
The model possesses the capacity to manage tens of thousands of lines of code and sustain coding efforts over multiple hours. On the GDPval-AA benchmark, Sonnet 5.5 scored 1,844, closely approaching the 1,846 score of Opus 5.5 and exceeding Sonnet 5’s 1,449.
Visual and general performance metrics also improved, with Sonnet 5.5 scoring 80.1% on computer-use performance and 61.6% on visual chart recognition. Notably, Sonnet 5.5 represents the first Sonnet model capable of beating Pokémon Red using only screenshots for input.
Security features are also enhanced, as Sonnet 5.5 is the first Sonnet model to incorporate cybersecurity safeguards and fallbacks similar to those used for Anthropic’s more advanced models. Availability for the public is confirmed, with Claude Sonnet 5.5 accessible across Amazon Web Services, Google Cloud, and Microsoft Azure.
Looking ahead, Claude Haiku 5.5 is scheduled to join the Claude 5.5 family within the coming weeks.
