Anthropic says Claude Sonnet 5.5 is 30% faster at same price

Anthropic released Claude Sonnet 5.5 on Monday, the second model in its Claude 5.5 family after Opus 5.5 last week. Anthropic says the new model “generates outputs 30%+ faster than Sonnet 5” and, in its own testing, costs “up to 30% less per task.”

What’s new

• Price: Unchanged from Sonnet 5 at $2 per million input tokens and $10 per million output tokens. Cache reads cost $0.20 per million tokens. Anthropic says the model “typically needs far fewer tokens to do the same work,” but gives no token percentage.
• Benchmark: Anthropic says Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0, an agentic coding test, compared with 10.3% for Sonnet 5. Anthropic’s system card lists Opus 5.5 at 66.4%. According to the card, Sonnet 5.5 ran in Claude Code at max effort with safeguards on, averaged over five trials per task, while Opus 5.5 ran at a different effort setting (“xhigh”). The card puts the standard error at about ±2.5 points. None of these results has been independently tested.
• Cyber: Anthropic says Sonnet 5.5’s cybersecurity abilities are comparable to Opus 5’s, making it the first Sonnet to launch with cyber safeguards like those on its most capable models. The system card describes three stages: a probe on the model’s internal activations, a lightweight classifier and a separate LLM classifier. Higher-risk cyber requests that get blocked fall back to Sonnet 5, automatically in Anthropic’s own apps and for API developers who opt in. Anthropic says users “should expect increased refusals,” even on benign security work.
• Availability: Anthropic says the model is live on all platforms, including AWS, Google Cloud and Microsoft Azure, as claude-sonnet-5-5.

What’s next

Refusal rates for security teams. Anthropic says verified cyberdefenders will “soon” be able to apply for an expanded Cyber Verification Program with access to more advanced capabilities. Claude Haiku 5.5 is due “in the coming weeks,” Anthropic says.

— Lark Pemberton