Anthropic: Claude Sonnet Latest

Anthropic: Claude Sonnet Latest is a rolling provider alias (~anthropic) that tracks the newest Claude Sonnet snapshot on the Claude API and partner platforms. As of late 2026 it resolves to Claude Sonnet 5.5, released on September 28, 2026 as the speed-optimized tier of the Claude 5.5 family. Sonnet 5.5 is a proprietary hybrid reasoning transformer with adaptive thinking, a 1M-token context window, up to 128K tokens of output, and text-plus-image input. It is positioned between Haiku (volume/latency) and Opus (deepest judgment) for agentic coding, terminal workflows, knowledge work, and multimodal tasks.

Relative to Claude Sonnet 5, Anthropic reports large gains on agentic coding and computer-use benchmarks while keeping list pricing at $2 per million input tokens and $10 per million output tokens. The Sonnet 5.5 system card and launch materials highlight strong scores on SWE-Bench Pro (81.3%), SWE-Bench Multilingual (90.3%), DeepSWE (71.0%), Humanity's Last Exam with tools (64.5%), GDPval-AA v2.1 (1844 Elo), AutomationBench (44.7%), and Toolathlon-Verified (77.8% Pass@1), with competitive multimodal chart understanding and OSWorld-style computer use. The model ships with expanded safety measures for cybersecurity and distillation resistance appropriate to its capability tier.

Claude Sonnet Latest is API-only (no public weights). Parameter count and training compute are undisclosed. Developers invoke it via the current Sonnet model ID (for example claude-sonnet-5-5 while that generation is current). Effort settings trade latency and cost against reasoning depth; the alias automatically advances when Anthropic promotes a newer Sonnet release.