<MCP.Playground/>
online

Claude Opus 5.5

Runs MCP toolsReasoning

Anthropic’s newest flagship, released 22 September 2026. Claude is the family MCP was built around, so Opus 5.5 is the baseline most servers are measured against. Run it on your own tools before you compare anything else.

Vendor
Anthropic
Released
22 September 2026
Context
1M tokens
Input
Text, image, file

Paste a server URL, pick a model, and watch it call your tools in a real conversation. No install.

No server of your own? Leave it blank and use one of the public mock servers.

In your browser

MCP Agent Studio

Server has a public URL? Chat with Claude Opus 5.5 on it, watch every tool call, and compare models side by side.

Open MCP Agent Studio →

On your computer

MCP Playground Desktop

Server runs on localhost or over stdio? The browser can’t reach it. The desktop app can, with Claude Opus 5.5 in Agent Studio. Download free

Mac (Apple Silicon)Mac (Intel)Windowsor Microsoft Store

MCP Evals

Can Claude Opus 5.5 actually complete tasks with your server?

One chat shows the tool calls work. An eval writes real tasks from your tool schemas, has Claude Opus 5.5 drive each one, and reports where it got the wrong answer even though every call succeeded. Pick Claude Opus 5.5 as the driver.

Run evals →See a sample

What Claude Opus 5.5 is

Claude Opus 5.5 succeeds Opus 5 as Anthropic’s flagship for demanding reasoning, coding and long-horizon agent work. Anthropic says the model matches Claude Fable 5.1 on most work and generates output more than 30% faster than Opus 5. It has a 1M-token window and accepts text, image and file input.

What Anthropic says

  • Terminal-Bench 4.0: 66.4%, against 55.8% for Fable 5.1.
  • AutomationBench: 40% task completion, against 26.9% for Opus 5.
  • Humanity’s Last Exam with tools: 67.7%.
  • Output generation more than 30% faster than Opus 5.

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, with no translation layer. On the API the model id is claude-opus-5-5.
  • Through Claude Code, Claude Desktop and claude.ai, all of which act as MCP clients
  • MCP Playground’s Agent Studio: paste your server URL above and run Opus 5.5 against it in the browser, with no Anthropic key

What to watch for

  • Anthropic says it matches Fable 5.1 on most work. If that holds for your tools, you may not need Fable for agent runs. Test both on the same prompt before you decide.
  • Because MCP’s conventions came from Claude, a server that only works on Opus may be relying on Claude-specific leniency. Passing on Opus 5.5 is a baseline, not proof that the server works everywhere.
  • On OpenRouter it accepts temperature, stop and verbosity, which the GPT-6 models do not. A client tuned on Claude can send parameters that GPT-6 silently ignores.

Run the error-simulation mock and ask Opus 5.5 to finish a task that hits simulate_partial_success. What matters is whether it reports the partial result honestly or retries until it can claim success. That is where long-horizon agents differ most. All the mock servers →

Available here

ModelModel ID
Claude Opus 5.5anthropic/claude-opus-5.5

Go deeper

Frequently asked questions

Is Claude Opus 6 available?
No. As of September 2026, Anthropic’s newest Opus is Opus 5.5, released on 22 September. There is no Opus 6 on Anthropic’s API or in Claude’s apps.
What changed from Claude Opus 5?
Anthropic reports faster output (more than 30%), AutomationBench task completion up from 26.9% to 40%, and attempts to circumvent containment boundaries down 85%. The window stays at 1M tokens.
Does Claude Opus 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.
Opus 5.5 or Fable 5.1 for agent work?
Anthropic says Opus 5.5 matches Fable 5.1 on most work and beats it on Terminal-Bench 4.0 (66.4% against 55.8%). Run both on your own tools. Whether the gap holds on your server is the question that matters.

Sources