Claude Sonnet 5.5: Same API Rates, Lower Task Costs

Claude Sonnet 5.5 keeps the same per-token API rates as Sonnet 5, while Anthropic says it can cut the cost of completing a task by up to 30%. The distinction matters: the claimed savings come from using fewer tokens and tool calls, not from a lower price for each token.\n\n## The efficiency claim is about the whole task\n\nAnthropic says Sonnet 5.5 generates output more than 30% faster than Sonnet 5. That speed figure is the company’s claim; VentureBeat reported it had asked Anthropic how the improvement was measured and was awaiting a response at publication.\n\nThe model is aimed at defined, everyday work such as coding, debugging, documents, spreadsheets and interface design. Anthropic’s cost argument is that fewer tokens and fewer agent tool calls can reduce the total work required to finish those jobs. At Box, for example, an executive said the model was 2.4 times faster and used 12% fewer total tokens than its predecessor in the company’s testing. These are customer-reported results, not a promise that every deployment will see the same gains. VentureBeat’s account of Anthropic’s claims and customer tests\n\nThat is a more useful way to think about agent economics than token rates alone. An agent that needs fewer retries or external actions may cost less to run even when the price per token is unchanged. But Anthropic’s “up to 30%” figure is a maximum claim, not a universal saving or an independently established production result.\n\n## Same API price, closer to Opus on selected tests\n\nSonnet 5.5 costs $2 per million input tokens and $10 per million output tokens, the same listed rates as Sonnet 5. Opus 5.5 costs $4 and $20, respectively. Anthropic also lists Sonnet cache reads at $0.20 per million tokens and cache writes at $2.50 per million. The reported API rates and benchmark results\n\nAnthropic reports Sonnet 5.5 scoring 70.6% on Terminal-Bench 4.0, compared with 66.4% for Opus 5.5 and 10.3% for Sonnet 5 under the reported settings. On GDPval-AA, which evaluates tasks drawn from real-world occupations, Sonnet 5.5 scored 1844, close to Opus 5.5’s 1846. Those results suggest the mid-tier model has narrowed the gap on some tests; they do not establish that it will match Opus on every job. Anthropic still positions Opus for difficult, ambiguous work that requires sustained judgment.\n\n## Access is broad, with a cyber-workflow exception\n\nDevelopers can access Sonnet 5.5 through the Claude Platform and Amazon Web Services, Google Cloud and Microsoft Azure. The Claude Platform model identifier is claude-sonnet-5-5. Availability across the listed platforms\n\nAnthropic also says Sonnet 5.5 is the first Sonnet model with cybersecurity safeguards modeled on those used for its higher-capability models. Routine software development and vulnerability remediation should continue to work normally, but some higher-risk cybersecurity requests may fall back to Sonnet 5. That distinction matters for security teams: broader availability does not mean every cyber-related request will be handled by the new model. Anthropic’s stated safeguards and fallback behavior\n\nThe practical question is therefore not simply whether Sonnet 5.5 has a lower API price. It does not. Its value depends on whether fewer steps and tokens make a particular workflow cheaper to complete, while its cyber safeguards may route some specialized work elsewhere.

FAQs

Anthropic says Sonnet 5.5 generates output more than 30% faster. The measurement method was not explained in the reporting, so treat this as a company claim rather than a guaranteed result.