Claude Opus 5.5: what's new and how to set its effort

Anthropic's Opus for long-running agentic coding and knowledge work. Cheaper and faster than Opus 5, with thinking that's always on and a lower default effort.

Model ID: claude-opus-5-5
Context: 1M
Max output: 128K
Price / MTok: $4 / $20
Default effort: medium

What it's good at

Built for long agentic coding sessions and knowledge work.
About 40% cheaper to run than Opus 5 for typical token-billed work, and more than 30% faster output, per Anthropic.
At its default effort (medium) it scores 52.5% on CursorBench, above Opus 5 at max (46.6%).
Reads dense charts, diagrams and screenshots much more precisely without tools.

What changed

Thinking is always on. Turning it off, or setting a manual thinking budget, returns an error. Effort is the only control.
The default effort is medium, one step below Opus 5's high. It also thinks more per turn at the same level.
Forced tool use (tool_choice any / tool) returns an error; use auto with strict tools.
Text between tool calls comes back as thinking blocks, empty unless you ask for them.

Which effort to use

Opus 5.5 defaults to medium, and it's a strong medium: its CursorBench score at medium beats Opus 5 at max. Start there. Raise to xhigh for complex refactors and long agentic runs; keep max for the hard bug you've been stuck on.

How to set effort in Claude Code

low

Fastest, fewest tokens. Simple edits, quick questions, subagents.

medium

Everyday work. The default on Opus 5.5, and on Sonnet 5.5 in Claude Code.

high

More reasoning for multi-file changes. The default on most models.

xhigh

Complex coding and long agentic runs. Deeper than high, without the full cost of max.

max

Hard debugging and architecture decisions. Session-only in Claude Code.

Five ways to set it

/effort

Open the slider. Or /effort xhigh to set a level, /effort auto to go back to the model's default.

/model

Use the left and right arrow keys on the effort slider while picking a model.

claude --effort high

One run only. The level doesn't carry into your next session.

"effortLevel": "xhigh"

In your settings file, to make it your standing default.

CLAUDE_CODE_EFFORT_LEVEL=high

As an environment variable (a level name, or auto).

A level set with /effort carries into your next session; max doesn't, unless you set it through the environment variable. If Claude skipped a file or stopped halfway, raise the effort. If it had everything and still got it wrong, try a more capable model.

Questions

What is the default effort for Claude Opus 5.5?

Medium, on the API and in Claude Code. Claude Opus 5 defaulted to high, so if you moved over and things feel shallower, set effort explicitly.

Can I turn thinking off on Opus 5.5?

No. Thinking is always on; a request that disables it returns a 400. Lower the effort instead. Low is the cheapest, fastest setting.

How much does Claude Opus 5.5 cost?

$4 per million input tokens and $20 per million output tokens, with cache reads at $0.20. Fast mode on the Claude API is $8 / $40.

Opus 5.5 or Sonnet 5.5?

Opus 5.5 for complex work that needs careful judgment and long agentic sessions; Sonnet 5.5 for well-scoped everyday tasks, bug fixes and documents, at half the price.

Sources

Use Claude in Claude Code today

A day pass — no Anthropic account, no waiting for your limit.

Get a Claude day pass