Claude Sonnet 5.5
Runs MCP toolsReasoning
Anthropic’s Sonnet-class model for everyday work, released 28 September 2026, six days after Opus 5.5. It is a direct upgrade to Sonnet 5 and scores within a few points of Opus 5.5 on most of Anthropic’s published benchmarks.
What Claude Sonnet 5.5 is
Claude Sonnet 5.5 succeeds Sonnet 5. Anthropic positions it as the faster, cheaper complement to Opus 5.5, aimed at well-scoped work: building features, fixing bugs, and document work. It has a 1M-token window, takes text, image and file input, and has five effort levels from low to max, with adaptive thinking on by default. On Anthropic’s API it is claude-sonnet-5-5.
What Anthropic says
- Terminal-Bench 4.0: 70.6%, above Opus 5.5’s 66.4%.
- OSWorld 2.1: 80.1%.
- GDPval-AA v2.1: 1844, against 1846 for Opus 5.5.
- CursorBench 4.0: 55.5%, about two points below Opus 5.5.
- Output generation 30%+ faster than Sonnet 5, by Anthropic’s account.
These are the vendor’s own claims, not measurements of ours. Run the model against your own server to find out whether they hold for your tools.
How it reaches MCP
- Natively. Anthropic’s Messages API can connect to remote MCP servers itself. On the API the model id is claude-sonnet-5-5.
- Through Claude Code, Claude Desktop and claude.ai, all of which are MCP clients
- MCP Playground’s Agent Studio: paste your server URL above and run Sonnet 5.5 against it in the browser, with no Anthropic key
What to watch for
- The default effort depends on where you run it. Anthropic sets Claude Code and the Claude apps to medium and the API to high. A tool sequence that works in Claude Code can take a different path through the API, so test at the effort your users will get.
- OpenRouter lists temperature for Sonnet 5.5, which it did not for Sonnet 5, but still no seed. Lowering temperature narrows the variation between runs. It does not make a tool call repeatable.
- It supports structured_outputs, so strict JSON-schema arguments are available. A tool schema that strict mode rejects shows up as a schema error, not a bad call.
Run the complex-schema mock on Sonnet 5 and Sonnet 5.5 with the same prompt, then on Opus 5.5. process_order has the deepest nesting of the four tools, so it is where an upgrade in argument handling shows first. All the mock servers →
Available here
| Model | Model ID |
|---|
| Claude Sonnet 5.5 | anthropic/claude-sonnet-5.5 |
Frequently asked questions
Does Claude Sonnet 5.5 support MCP?
Yes, natively. Anthropic’s API can connect to remote MCP servers directly, and Claude Code, Claude Desktop and claude.ai are all MCP clients. You can also run it against any server here without an Anthropic key.
Claude Sonnet 5.5 or Opus 5.5 for MCP tools?
Start with Sonnet 5.5. Anthropic puts it within a few points of Opus 5.5 on most benchmarks and ahead of it on Terminal-Bench 4.0. Run the same prompt on both against your server here. If the calls match, Sonnet is the one to ship.
What changed from Claude Sonnet 5?
Anthropic calls it a direct upgrade. The biggest published jump is Terminal-Bench 4.0, from 10.3% to 70.6%, and output is 30%+ faster. Both are available here if you want to compare them on your own tools.
When was Claude Sonnet 5.5 released?
28 September 2026, six days after Claude Opus 5.5.