Claude Sonnet 5.5 promises Opus-level results without the premium price

Anthropic has released Claude Sonnet 5.5, a new version of its midrange AI model that the company says is faster, more efficient, and substantially stronger at coding and everyday knowledge work than Claude Sonnet 5. Anthropic announces in an official blog post that Sonnet 5.5 is available across its own platform and through Amazon Web Services, Google Cloud, and Microsoft Azure.

The company positions the model as a lower-cost counterpart to Claude Opus 5.5. Opus remains intended for complex, open-ended assignments that require sustained judgment. Sonnet 5.5, by contrast, targets defined day-to-day tasks such as debugging software, producing documents and presentations, analyzing information, and building spreadsheets.

Anthropic says Sonnet 5.5 produces output more than 30% faster than Sonnet 5. Its listed token prices remain unchanged at $2 per million input tokens and $10 per million output tokens. However, the company says the newer model often needs fewer tokens and fewer tool calls to complete the same task, reducing total task costs by as much as 30% in its testing.

Large gains in coding benchmarks

The strongest claimed improvement concerns software development. On Terminal-Bench 4.0, an evaluation of multi-step command-line coding work, Sonnet 5.5 scored 70.6%, compared with 10.3% for Sonnet 5. Anthropic reports that this result exceeds the 66.4% score listed for Opus 5.5 at its highest effort setting, although the company cautions that benchmark results capture only part of a model’s practical performance.

On other coding tests, Sonnet 5.5 comes close to Opus 5.5 but does not consistently surpass it. It scored 55.5% on CursorBench 4.0, compared with 57.8% for Opus 5.5 and 34.1% for Sonnet 5. On FrontierCode 1.1, Sonnet 5.5 reached 52.1% at its highest effort setting, versus 54.4% for Opus 5.5.

Effort settings allow users to trade speed and cost for more extensive reasoning and checking. Anthropic says its consumer apps and Claude Code use Medium effort by default, while the Claude Platform defaults to High effort. The company argues that Sonnet 5.5 can outperform Sonnet 5 at lower effort levels while using a fraction of the cost per task.

Document work and safety controls

Anthropic also reports gains in knowledge work, including computer use, chart recognition, document analysis, and long-running assignments. On GDPval-AA, a benchmark covering work across 44 occupations, Sonnet 5.5 scored 1844, nearly matching Opus 5.5 at 1846 and exceeding Sonnet 5 at 1449. The company says early testers also found the model better at producing polished slides and interfaces.

The release includes tighter safeguards for some sensitive uses. Anthropic says Sonnet 5.5 has cybersecurity capabilities comparable to Opus 5 and will use similar protections. Requests judged to involve higher-risk cyber activity may fall back to Sonnet 5. Routine programming work should remain unaffected, according to the company.