Anthropic released the Claude Sonnet 5.5 model on September 28, 2026. It quickly secured third place on the Agent Arena leaderboard, which evaluates AI performance on multi-step agentic tasks. The model achieved a net improvement score of 12.52%, with a margin of error of plus or minus 3.09%.
The leaderboard currently ranks Claude Fable 5.1 at 14.31% and Claude Opus 5.5 at 13.82% in the top two spots. OpenAI’s GPT-6 Astra follows in fourth place with a score of 12.27%. The difference between the top four models is smaller than the margin of error for the Sonnet 5.5 score.
The model features a context window of 1 million tokens and a maximum output of 128,000 tokens. Pricing remains unchanged at $2 per million input tokens and $10 per million output tokens. However, the median cost per task is estimated at approximately $2.78, influenced by higher token usage compared to peers.
In other benchmarks, Sonnet 5.5 reached a score of 56 on the Artificial Analysis Intelligence Index, placing it second behind Opus 5.5. On the Terminal-Bench 4.0 test, the model recorded a result of 70.6%, significantly higher than the 10 to 14% range achieved by its predecessor, Sonnet 5.
