Anthropic ships Sonnet 5.5: 30% faster, up to 30% cheaper per task

赛博禅心 · wechat · 2026-09-29

A detailed Chinese write-up of Anthropic's overnight release of Claude Sonnet 5.5. It keeps Sonnet 5 pricing ($2/M input, $10/M output tokens — half of Opus 5.5), runs 30%+ faster, and cuts per-task cost by up to 30% thanks to lower token usage. Anthropic positions Opus 5.5 for complex open-ended tasks and Sonnet 5.5 for well-scoped daily work, bug fixing, docs/slides; Haiku 5.5 arrives in weeks.

Benchmarks: on Terminal-Bench 4.0, best per-tier results cost under 1/10 of Sonnet 5's, and Max effort hits 70.6% to beat Opus 5.5; FrontierCode beats same-tier Sonnet 5 by 10 points; CursorBench 4.0 scores 55.5%, second only to Opus 5.5. Customer data: Base44 built 118 real apps with 3.6 iterations vs 7.7 for Opus 5; Zendesk handles tickets 20% faster; Box is faster and more accurate. It's also the first Sonnet to beat Pokémon Red from screenshots alone, and the first with cyber-safety guardrails. Live on all platforms as claude-sonnet-5-5.

Original post →

Anthropic ships Sonnet 5.5: 30% faster, up to 30% cheaper per task

More from Models