Anthropic released Claude Sonnet 5.5 on 2026-09-28, six days after Claude Opus 5.5, calling it “the second model in the Claude 5.5 family.” Anthropic says it is “a clear upgrade over Claude Sonnet 5, runs 30%+ faster, and costs up to 30% less for most work.” Token prices are unchanged from Sonnet 5 (2 dollars per million input tokens, 10 dollars per million output, 0.20 dollars for cache reads); the saving comes from needing fewer tokens per task. It is available on the Claude Platform as claude-sonnet-5-5 and on AWS, Google Cloud and Microsoft Azure. Anthropic says Claude Haiku 5.5 will follow “in the coming weeks.”
Anthropic’s published table shows Sonnet 5.5 close to Opus 5.5 on several tests and ahead of it on one: Terminal-Bench 4.0 at 70.6 percent (Opus 5.5: 66.4), CursorBench 4.0 at 55.5 percent (57.8), GDPval-AA v2.1 at 1844 Elo (1846, against 1449 for Sonnet 5), OSWorld 2.1 partial reward at 80.1 percent (81.8) and Humanity’s Last Exam with tools at 64.5 percent (67.7). Anthropic also says it is the first Sonnet model to beat Pokemon Red working only from screenshots. It adds that “Opus 5.5 remains clearly stronger at complex, open-ended work requiring sustained judgment.”
Because its cybersecurity capability is comparable to Opus 5’s, Sonnet 5.5 is “the first Sonnet model to launch with cyber safeguards and fallbacks like those we’ve developed for our most capable models”; its biology safeguards are unchanged from Sonnet 5. Developers who ran Sonnet with thinking off must switch to a new between_tools setting.
Why it matters: the mid-priced tier now posts benchmark numbers in the same range as the flagship released a week earlier, and the fact that a Sonnet model needs cyber gating shows how far capability has moved down the price ladder. What it does not show: all figures are Anthropic’s own, the “up to 30% less” cost claim depends on workload, and the comparisons with GPT-6 Sol omit some benchmarks where OpenAI had not published numbers.