Anthropic releases Claude Sonnet 5.5 near Opus at same price

Anthropic released Claude Sonnet 5.5 at Sonnet 5’s $2/$10 price, with a 70.6% Terminal-Bench score and claims of faster output and up to 30% lower task cost.

Anthropic released closed-weights Claude Sonnet 5.5 (claude-sonnet-5-5) about three months after Sonnet 5 as a faster, lower-cost model for everyday coding, bug fixes, and office documents, on the Claude Platform, AWS, Google Cloud, and Microsoft Azure, with zero data retention reported on those three clouds. Prices match Sonnet 5 at $2 per million input tokens and $10 per million output tokens, plus $0.20 cache reads, and Anthropic says fewer tokens and tool calls can cut cost per task by up to 30%. Most reports say output is more than 30% faster than Sonnet 5; TechCrunch says 30% faster and that it burns tokens more slowly. Cited 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 versus 1,846; TechCrunch says it leads Opus 5.5 on agentic coding, which matches Terminal-Bench but not CursorBench, while Anthropic says Opus 5.5 stays stronger on complex open-ended work. It is the first Sonnet under cyber safeguards used for Fable and Opus, with capability described as comparable to Opus 5, high-risk cyber requests rerouted to Sonnet 5, distillation-attack classifiers added, biology safeguards unchanged, and alignment at least matching Sonnet 5 in an audit of about 1,850 scenarios; a new Haiku is planned within weeks. Separately, Anthropic’s Opus 5.5 prompting guide says existing Opus 5 prompts should still work, medium effort is the default and can match or beat Opus 5 at high effort, thinking is always on inside a max_tokens cap up to 128,000, and tokens are more than 30% faster; Simon Willison also reported a Sonnet 5.5 run that hit that 128,000-token bug and failed after $1.28, while an xhigh run succeeded in 41 seconds for 5.74 cents.

• On 28–29 September 2026 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: $2 per million input tokens and $10 per million output tokens, with cache reads at $0.20; Anthropic claims up to 30% lower cost per task from fewer tokens and tool calls.
• Speed wording differs: several outlets say more than 30% (or 30%+) faster than Sonnet 5, while TechCrunch says 30% faster and slower token burn.
• Scores cited: Terminal-Bench 4.0 70.6% (Sonnet 5 10.3%; Opus 5.5 66.4%); CursorBench 4.0 55.5% vs Opus 5.5’s 57.8%; GDPval-AA 1,844 vs Opus 5.5’s 1,846 (MarkTechPost specifies GDPval-AA v2.1).
• MarkTechPost lists a 1 million-token context, 128K max output, and a June 2026 knowledge cutoff; Anthropic says Opus 5.5 remains stronger on complex open-ended work.
• First Sonnet under cyber safeguards used for Fable and Opus; cyber capability described as comparable to Opus 5; high-risk cyber requests reroute to Sonnet 5; biology safeguards unchanged from Sonnet 5; ~1,850-scenario audit found…
• A new Haiku (Haiku 5.5, per The Decoder) is planned in the coming weeks with no date; one report says it is the first Sonnet to finish Pokémon Red from screenshots only.
• Same day, Anthropic’s Opus 5.5 prompting guide says Opus 5 prompts should still work, effort defaults to medium, thinking is always on and counts toward max_tokens up to 128,000, and output is more than 30% faster than Opus 5.