Anthropic releases Claude Sonnet 5.5 near Opus at same price

Anthropic launched Claude Sonnet 5.5 at Sonnet 5's $2/$10 price, scoring 70.6% on Terminal-Bench 4.0, with cyber safeguards and up to 30% lower task cost.

Anthropic released closed-weights Claude Sonnet 5.5, model id claude-sonnet-5-5, on 28 September 2026, about three months after Sonnet 5, on the Claude Platform, AWS, Google Cloud, and Microsoft Azure. It keeps Sonnet 5 pricing of $2 per million input tokens and $10 per million output tokens, with The Decoder also citing $0.20 cache reads, while Anthropic says it is more than 30% faster and up to 30% cheaper per task because it uses fewer tokens and tool calls; TechCrunch instead summarized the gain as 30% faster with a slower token burn. Vendor scores are 70.6% on Terminal-Bench 4.0 versus 10.3% for Sonnet 5 and 66.4% for Opus 5.5, 55.5% on CursorBench 4.0 versus Opus 5.5's 57.8%, and 1,844 on GDPval-AA (called v2.1 by MarkTechPost) versus 1,846 for Opus 5.5, so it nearly matches Opus on that knowledge score while leading Opus on Terminal-Bench but trailing slightly on CursorBench, contrary to a blanket claim that it leads on agentic coding. Because cyber capability is described as comparable to Opus, it is the first Sonnet under the cyber safeguards used for Fable and Opus: high-risk cybersecurity requests fall back to Sonnet 5, distillation-attack classifiers were added, and biology safeguards stay as on Sonnet 5. MarkTechPost lists a 1 million-token context, 128K max output, and a June 2026 cutoff, and says Opus 5.5 remains stronger on complex open-ended work; an alignment audit of about 1,850 scenarios at least matched Sonnet 5, and a new Haiku 5.5 is planned in the coming weeks with no date.

• Anthropic released closed-weights Claude Sonnet 5.5 (claude-sonnet-5-5) about three months after Sonnet 5, on the Claude Platform, AWS, Google Cloud, and Microsoft Azure.
• List price matches Sonnet 5 at $2 per million input tokens and $10 per million output tokens; The Decoder also reports $0.20 cache reads. Anthropic says it is more than 30% faster and up to 30% cheaper per task via fewer tokens and tool…
• Terminal-Bench 4.0: 70.6%, versus 10.3% for Sonnet 5 and 66.4% for Opus 5.5. CursorBench 4.0: 55.5%, versus 57.8% for Opus 5.5.
• GDPval-AA (v2.1 per MarkTechPost): 1,844, two points below Opus 5.5 at 1,846.
• Specs reported by MarkTechPost: 1 million-token context, 128K maximum output, June 2026 knowledge cutoff; Opus 5.5 remains stronger on complex open-ended work.
• First Sonnet under the cyber safeguards used for Fable and Opus; high-risk cybersecurity requests fall back to Sonnet 5, distillation-attack classifiers were added, and biology safeguards are unchanged from Sonnet 5.
• An audit of about 1,850 scenarios found alignment at least matching Sonnet 5. Claude Haiku 5.5 is planned in the coming weeks, with no date.
• Simon Willison: a max-effort pelican SVG hit the same 128,000-token failure as Opus 5.5 and cost $1.28; an xhigh run succeeded in 41 seconds for 5.74 cents. Hacker News: first Sonnet to finish Pokémon Red from screenshots only.