Claude Sonnet 5.5 (EU)

Claude Sonnet 5.5 debuted at No. 2 on the Artificial Analysis Intelligence Index with 56 points, just two behind Anthropic's own Opus 5.5 at 58 and ahead of OpenAI's GPT-6 Astra at 53 and GPT-6 Sol at 48. The 18-point gain over Sonnet 5 (38 points) is the largest single-generation jump on the index. On Terminal-Bench 4.0, Sonnet 5.5 reaches 64%, ahead of both Opus 5.5 and GPT-6 Astra at 60%, a 50-point gain over Sonnet 5. On Terminal-Bench-Science, Sonnet 5.5 scores 53%, trailing only GPT-6 Astra and Opus 5.5. For classic knowledge work it essentially matches Opus 5.5 on GDPval-AA (1,844 vs 1,846 Elo), AA-Briefcase (1,811 vs 1,822), and AutomationBench-AA (71% vs 70%). On AA-Omniscience factual knowledge it answers 54% correctly versus 66% for Opus 5.5, but hallucinates less (47% vs 59%), reflecting a deliberate speed-and-efficiency trade-off.

Anthropic's Sonnet 5.5 is a more compact, efficiency-focused counterpart to Opus 5.5, built for shorter texts and lower-latency output while still approaching flagship capability. It scores 70.6% on Terminal-Bench 4.0 versus 10.3% for Sonnet 5, and 61.6% on a chart-reading test where Sonnet 5 reached only 15.6%. At "Max effort" the model approaches Opus 5.5 quality, making the jump from Sonnet to Opus a continuous trade-off rather than a step change. The Claude 5.5 generation is broader than 5.1 or 4.8, with Haiku 5.5 expected next; Opus 5.5 and Sonnet 5.5 were released after Anthropic publicly called for a pause on more capable models. Anthropic expressed confidence the new models exhibit acceptable and safe behavior, positioning the lineup as a coordinated release of efficiency- and capability-tier options across the same family.

Anthropic released Claude Sonnet 5.5 in late September 2026, positioning it as the second model in the Claude 5.5 series with a speed improvement of over 30% and a cost reduction of nearly 30% versus Sonnet 5. Pricing is held at $2 per million input tokens, $10 per million output tokens, and $0.20 per million tokens for cache reads, matching the previous generation despite a 30% lower per-task cost in practice. The model is available across all major platforms including AWS, Google Cloud, and Microsoft Azure. On the Artificial Analysis Intelligence Index, Sonnet 5.5 scored 56, second only to Opus 5.5 at 58, and far ahead of Sonnet 5 at 38. The model is described as the first Sonnet capable of completing Pokémon Red via screenshots, and a Rubik's Cube restoration demo ran in 10 seconds for $0.11. Developers reported that real measured per-task costs reached about $7.60, roughly 50% higher than Sonnet 5's per-task cost and close to Opus 5.5's $7.63, prompting some to recommend staying on Opus 5.5.