Anthropic launches Claude Sonnet 5.5: 30% faster and up to 30% cheaper per task
dl_weekly · x · 2026-10-04
Anthropic released Claude Sonnet 5.5, the second model in the Claude 5.5 family, positioned as a faster, cheaper complement to Opus 5.5 for well-scoped everyday work: bug fixes, polished docs, slides, and spreadsheets.
- Performance: scores 70.6% on Terminal-Bench 4.0 (vs. 10.3% for Sonnet 5), just 2 points below Opus 5.5 on GDPval-AA, and is the first Sonnet to beat Pokémon Red from screenshots alone
- Cost: same pricing as Sonnet 5 ($2/M input, $10/M output, $0.20/M cache reads) but needs far fewer tokens — up to 30% cheaper per task in testing
- Speed: generates outputs 30%+ faster, Anthropic's fastest Sonnet yet
- Collaboration: clearer writing; early testers preferred it over Sonnet 5
- Safety: matches or improves on Sonnet 5 across most alignment metrics; comparable cybersecurity capabilities
- Claude Haiku 5.5 for high-volume, cost-sensitive apps arrives in coming weeks
More from Models
- Yacine teases dragging a nonexistent 'Opus 5.5' out of distribution and forcing it to think — yacineMTB · 2026-10-04
- Jev probed on 16,379 benchmark requests: not frontier, but bends the price-performance curve — enn_nafnlaus · 2026-10-04
- GPT Sol is painfully slow, but accurate enough to 'trust with surgery' — Lonely_Potato70 · 2026-10-04
- Qwen 3.8 Flash Next local tests hit 80-110 tok/s, poised to replace the 27b model — inthesearchof · 2026-10-04
- OpenAI turned ChatGPT into a feature dumpster while Anthropic just ships better models — _AustinCalvert_ · 2026-10-04
- Gemma model caught in a 'doom loop': apologizes for misspelling, then keeps repeating it — adariostrange · 2026-10-04