Claude Sonnet 4.6 (JP)

Anthropic's official Claude Sonnet 4.6 System Card, dated February 17, 2026, documents the model as Anthropic's latest large language model and describes evaluations across coding, agentic, reasoning, multimodal, computer use, and mathematical abilities, alongside domain-specific assessments in finance, cybersecurity, The system card outlines an alignment assessment covering a wide range of potentially misaligned behaviors and tests in unusual and extreme scenarios, concluding that Sonnet 4.6 shows low overall misaligned behavior and, on some measures, the best alignment seen in any Claude model. The model was deployed under the AI

A third-party benchmark registry, benchmarks.company, compiles sourced and dated capability scores for Anthropic's Claude Sonnet 4.6, listing developer-reported results including SWE-bench Verified 79.6%, SWE-bench Pro 58.1%, Terminal-Bench 2.0 59.1%, Terminal-Bench 2.1 67%, tau-bench 97.9%, GPQA Diamond 89.9%, Humanit As an independent aggregation point, the page provides a head-to-head comparison feature against other registered models, supporting cross-model capability analysis of Claude Sonnet 4.6. Note that this evidence covers the Sonnet 4.6 family at large and does not explicitly address the jp.anthropic.claude-sonnet-4-6 Bedr