Anthropic launches Claude Sonnet 5.5 with new cyber safeguards

Claude Sonnet 5.5 is now available, replacing Sonnet 5 as Anthropic’s mid-tier model. The company says it generates output more than 30% faster and can cost up to 30% less per task, while its per-token rates stay unchanged. Its launch also brings cyber safeguards and distillation protections to the Sonnet line. Anthropic’s announcement sets out the model’s pricing, benchmarks and safety changes.

TechCrunch’s launch report describes Sonnet 5.5 as the company’s mid-tier upgrade, with speed and lower task costs as its main pitch. The headline numbers come from Anthropic’s own evaluations, so they are best read as the company’s comparison rather than an independent guarantee for every workload.

Claude Sonnet 5.5 posts a sharp coding benchmark gain

On Terminal-Bench 4.0, an evaluation of multi-step work in a command-line environment, Anthropic reports a 70.6% score for Sonnet 5.5, compared with 10.3% for Sonnet 5. The same table lists Opus 5.5 at 66.4%, with that Opus result measured at Xhigh effort, its highest setting.

¹ Anthropic reports Opus 5.5’s Terminal-Bench result at Xhigh effort. Scores are from the company’s published comparison.

The figures put Sonnet 5.5 ahead of its predecessor on both coding tests and slightly above Opus 5.5 on Terminal-Bench, but not on CursorBench. Anthropic says Opus remains stronger at complex, open-ended work that requires sustained judgement. That leaves Sonnet positioned for bounded jobs such as debugging, document creation and fast iteration, rather than as a blanket replacement for the higher-tier model.

The launch page also says Sonnet 5.5 is the first Sonnet to complete Pokémon Red using screenshots alone. That is a long-horizon task demonstration, not a standard measure of coding quality, and Anthropic’s broader benchmark results still need to be separated from claims about individual use cases.

Claude Sonnet 5.5 keeps token prices flat

Anthropic lists the same per-token rates as Sonnet 5: $2 per million input tokens and $10 per million output tokens. Cache reads cost $0.20 per million tokens. Opus 5.5 is listed at $4 for input and $20 for output, putting Sonnet at half the flagship’s token price. Tbreak covered Opus 5.5’s launch and pricing last week.

The lower-cost claim concerns the bill for a completed task, not the price of each token. Anthropic says Sonnet 5.5 typically uses fewer tokens and costs up to 30% less on its tested work. Decrypt’s report, citing Artificial Analysis, says an independent evaluation found the model used more tokens per task than any other model in that tester’s measured set. The two findings use different workloads; actual spend will depend on the task, output length and effort setting.

Claude Sonnet 5.5 adds cyber and distillation safeguards

Anthropic says Sonnet 5.5’s cybersecurity capability has improved substantially over Sonnet 5. Higher-risk cybersecurity requests will visibly fall back to Sonnet 5, while routine software development, including finding and fixing bugs, remains available. Cyber defenders will be able to apply for broader access through Anthropic’s Cyber Verification Program.

The model keeps the same biology safeguards as Sonnet 5. Anthropic says those controls target high-risk requests, while routine life-sciences work is generally unaffected. Separately, Sonnet 5.5 is the first Sonnet model to use safety classifiers intended to prevent distillation attacks that extract a model’s reasoning at scale. The release also expands preserved thinking, tying that reasoning to the account that created it.

Where Claude Sonnet 5.5 is available

Sonnet 5.5 is available through Claude, the Claude API and major cloud platforms including Amazon Web Services, Google Cloud and Microsoft Azure. Developers can call it with the model ID claude-sonnet-5-5. In Claude Code, the sonnet alias switches to 5.5 on version 2.1.284 or later; Opus 5.5 remains the default, so users need to select Sonnet explicitly, according to Anthropic’s developer guide. For integrations that disable upfront thinking, Anthropic’s migration guide says to use the new between_tools setting. The new model arrives after Anthropic added a coordinator for parallel work in Claude Code Projects.

Anthropic’s launch materials quote API rates in US dollars and do not list a UAE-specific price. Claude Haiku 5.5, the smaller model aimed at high-volume and cost-sensitive work, is still due in the coming weeks.