On September 28, 2026, Anthropic introduced Claude Sonnet 5.5 for everyday work, coding and document creation. The company says the new model generates output more than 30% faster than Sonnet 5 and can reduce cost per completed task by up to 30%, despite unchanged token prices. That distinction matters when teams evaluate multistep AI workflows.
Sonnet 5.5 complements Opus 5.5. Anthropic positions Sonnet for well-scoped work, including bug fixes, interface iteration, spreadsheets, presentations and documents, while Opus targets more open-ended problems requiring sustained judgment. Those descriptions reflect the vendor's positioning, not guaranteed outcomes in every deployment.
Anthropic's official announcement contains the original cost-versus-performance charts and side-by-side model demonstrations. The images remain on the company's website so readers can view them in context and without unauthorized redistribution.
See the original charts at Anthropic ↗Source: Anthropic, September 28, 2026. AUXER does not reproduce or claim ownership of the graphics.
What Anthropic reported in coding benchmarks
On Terminal-Bench 4.0, an agentic coding evaluation involving multistep terminal tasks, Anthropic reported 70.6% for Sonnet 5.5 and 10.3% for Sonnet 5. Its published CursorBench 4.0 results were 55.5% for Sonnet 5.5 and 34.1% for Sonnet 5. These are results for specific evaluation conditions, not universal success rates for software development.
| Evaluation | Sonnet 5.5 | Sonnet 5 |
|---|---|---|
| Terminal-Bench 4.0 | 70.6% | 10.3% |
| CursorBench 4.0 | 55.5% | 34.1% |
Vendor-reported results published September 28, 2026. Benchmark scores provide comparisons under defined conditions, not predictions for every project.
Token prices remain unchanged
Sonnet 5.5 costs $2 per million input tokens, $10 per million output tokens and $0.20 per million cached input tokens read. Anthropic also lists $2.50 per million tokens written to the five-minute cache. The input and output prices match the listed Sonnet 5 rates.
Anthropic's task-cost claim rests on completing certain jobs with fewer tokens and steps. Actual savings depend on effort settings, tool calls and retries. For a coding agent, the most useful measure is not the cost of one response; it is the cost of a tested, completed fix.
What developers should check before migrating
The API model ID is claude-sonnet-5-5. Official documentation lists a one-million-token context window and a standard maximum output of 128,000 tokens. It also identifies migration considerations involving thinking settings, sampling parameters and response handling.
Compare the models on real examples: debugging, tool-based tasks, structured output and review workflows. Track quality, cost per completed task, end-to-end latency and human intervention. For sensitive operations, continue to enforce permissions and checks outside the model itself.
Official sources
- Anthropic: Claude Sonnet 5.5, official announcement, evaluation tables and original charts, September 28, 2026.
- Claude Platform: Sonnet 5.5 documentation, specifications, pricing, API model ID and migration notes.