Anthropic · 2026-09-28 · seismic
Claude Sonnet 5.5 — Anthropic's faster Sonnet tops Opus 5.5 on Terminal-Bench
Claude Sonnet 5.5 is Anthropic's new mid-tier model. It runs over 30% faster than Sonnet 5 at the same $2 / $10 per million token price, and scores 70.6% on Terminal-Bench 4.0, ahead of Claude Opus 5.5.
Anthropic's mid-tier model gets faster and cheaper per task, and edges past Opus 5.5 on one agentic coding test.
Quick facts
| Maker | Anthropic |
|---|---|
| Model ID | claude-sonnet-5-5 |
| Context window | 1M tokens |
| Max output | 128K tokens |
| Knowledge cutoff | Jun 2026 |
| Default effort | high |
| Availability | Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
Benchmarks
Pricing
| Input | $2.00 / 1M tokens |
|---|---|
| Output | $10.00 / 1M tokens |
| Cache read | $0.20 / 1M tokens |
| Cache write (5m) | $2.50 / 1M tokens |
| Batch API · input and output | 50% off |
What is it?
Claude Sonnet 5.5 is the Sonnet-tier member of Anthropic's Claude 5.5 family, released September 28, 2026, six days after Claude Opus 5.5. Anthropic says it generates output more than 30% faster than Sonnet 5 and costs up to 30% less per task, while the per-token price stays at $2 input and $10 output. It is also the first Sonnet model to launch with the cyber safeguards and fallbacks Anthropic built for its most capable models.
How does it work?
Adaptive thinking is on by default and its depth is set with an `effort` parameter that defaults to `high`. A new lowest setting, `between_tools`, replaces `disabled` and turns off up-front thinking. Five breaking changes come with the upgrade: `disabled` thinking and forced tool use return errors, thinking blocks are tied to the model and conversation, the older `computer_20251124` tool is rejected on the Claude API and Google Cloud, and some advisor pairings no longer work. Requests flagged as higher-risk cyber work fall back to Sonnet 5.
Why does it matter?
On Anthropic's own tests the mid-tier model now sits close to the flagship: 70.6% on Terminal-Bench 4.0 against 66.4% for Opus 5.5, and 80.1% on OSWorld 2.1 against 81.8%. For teams running coding and computer-use agents at volume, that is near-flagship work at half the Opus 5.5 token price. Early HN reactions are mixed, with some developers reporting that the new cyber safeguards trigger on routine tasks.
Who is it for?
teams running coding agents and high-volume API workloads
Frequently asked questions
- How much does Claude Sonnet 5.5 cost?
- Claude Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens, the same as Claude Sonnet 5. Cache reads are $0.20 per million and a 5-minute cache write is $2.50 per million. Batch API requests take 50% off. Anthropic says Sonnet 5.5 still costs up to 30% less per task because it uses fewer tokens.
- How does Claude Sonnet 5.5 compare to Claude Opus 5.5?
- Claude Sonnet 5.5 beats Claude Opus 5.5 on Terminal-Bench 4.0, 70.6% to 66.4%, and trails it narrowly elsewhere: 80.1% to 81.8% on OSWorld 2.1, 55.5% to 57.8% on CursorBench 4.0 and 64.5% to 67.7% on Humanity's Last Exam. Sonnet 5.5 costs $2 / $10 per million tokens against $4 / $20 for Opus 5.5.
- What breaks when upgrading from Claude Sonnet 5 to Claude Sonnet 5.5?
- Five changes break Claude Sonnet 5 code. `thinking: disabled` returns an error and must become `between_tools`, forced tool use is rejected, thinking blocks are tied to the model and conversation, the `computer_20251124` tool is refused on the Claude API and Google Cloud, and Opus 4.8, Opus 4.7 and Sonnet 5 no longer work as advisors. Anthropic publishes a migration guide.
- Why does Claude Sonnet 5.5 sometimes fall back to Claude Sonnet 5?
- Claude Sonnet 5.5 is the first Sonnet model to ship with the cyber safeguards Anthropic uses on its most capable models. When a request is classed as higher-risk cyber work, it can fall back to Claude Sonnet 5 instead. Anthropic says cyber defenders can apply to its expanded Cyber Verification Program.
Try it
Model ID `claude-sonnet-5-5` on the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry