Opus 5.5 in Claude Code: Fable-level, faster, and cheaper than Opus 5

TL;DR Opus 5.5 beats Fable 5.1 on all nine benchmarks Anthropic published (66.4% against 55.8% on Terminal-Bench 4.0), generates output over 30% faster than Opus 5, and costs $4/$20 per MTok against $5/$25. It's already the Default on every plan, Pro included. Two conditions to get all of it: it needs Claude Code v2.1.280, and it starts at medium even if your settings.json says high. One claude-opus-5-5 entry under modelSettings fixes that.

Opus 5.5 shipped today, September 22, 2026, the first model in the Claude 5.5 family. Anthropic's one-line pitch: it performs at Fable 5.1's level on most work and costs 40% less than Opus 5. In Claude Code it becomes the default model.

What it brings

Fable-level coding

These are Anthropic's numbers, not mine. In their table it beats Fable 5.1 on all nine benchmarks, and on the three agentic coding ones by a clear margin.

Faster

It generates output more than 30% faster than Opus 5. And it needs fewer steps to get there: the launch note cites more terminal tasks solved than Opus 5 in less than half the steps. Fewer steps means less time waiting and fewer output tokens.

Cheaper than Opus 5

Per token it's 20% less than Opus 5; the rest of Anthropic's 40% comes from spending fewer tokens per task. Look at the last row too: cache reads are the line that weighs most in a long session (I measured it on Fable 5.1), and there Opus 5.5 is the cheapest of the three.

On a subscription you never see the per-token price. What you notice is the speed and each task using less. Anthropic is also raising the five-hour limits on Pro, Max, Team, and seat-based Enterprise.

My first impression, after half an hour and with all the caution that deserves: it answers faster and sounds more natural, which Anthropic highlights too (it leads with what matters and follows the style rules you give it).

It's your Default now, Pro included

With Opus 5, the Default on Pro and Team Standard stayed on Sonnet 5. Version 2.1.280 changes that: default now resolves to Opus 5.5 on Pro, Max, Team, Enterprise, and the API.

If you're on Pro and preferred Sonnet to make your five-hour limit last, you now have to ask for it: /model sonnet in the session, or "model": "sonnet" in your settings.json.

Before you rely on it, two checks

Two things decide whether you're really running it, and both fail silently.

1. If it isn't showing up, it's your version

Opus 5.5 requires Claude Code v2.1.280. Earlier today I was on 2.1.278, and the opus alias kept answering with Opus 5 without a word:

claude --model opus -p "Reply with just: ok" --output-format json
# "modelUsage": { "claude-opus-5": { ... } }

Ask for it by full name and at least you get told:

$ claude --model claude-opus-5-5 -p "Reply with just: ok"
API Error: 400 Claude Code 2.1.278 does not support this model; version 2.1.280 or newer is required. Run 'claude update', or update the Claude desktop app, then try again.

The fix is claude update. Afterwards the same command returns claude-opus-5-5. Two setups where that isn't enough today:

• Homebrew. The claude-code cask tracks the stable channel and sits at 2.1.267; claude-code@latest sits at 2.1.278. Neither has reached 2.1.280 yet, and Homebrew doesn't auto-update. If that's how you installed, you wait for the cask to move or switch to the native installer.
• Stable channel. With "autoUpdatesChannel": "stable" you get releases roughly a week late. That channel is on 2.1.267 today.

The Claude desktop app was already running 2.1.280 on my Mac.

2. It starts at medium, and your global effort doesn't reach it

Opus 5 defaults to high effort. Opus 5.5 defaults to medium.

That wouldn't matter if your own setting won, but it doesn't. The top-level effortLevel key in settings.json is the old form, the one /effort wrote before Claude Code started saving a level per model. It still applies to Opus 5, Fable 5.1, and older models. Not to Opus 5.5, and not to anything released after it. This only affects the key in your user settings.json: a top-level effortLevel in project, local, or managed settings still applies to every model.

Part of my settings.json:

{
  "effortLevel": "high",
  "modelSettings": {
    "claude-opus-5": { "effortLevel": "high" },
    "claude-fable-5-1": { "effortLevel": "high" }
  }
}

To read the real level I use the six-line skill from the effort chain, saved as .claude/skills/whicheffort/SKILL.md:

---
name: whicheffort
description: Report the active effort level
---
Reply with exactly this line and nothing else:

EFFORT=${CLAUDE_EFFORT}

On 2.1.280:

$ claude -p "/whicheffort"
EFFORT=medium

$ claude --model claude-opus-5 -p "/whicheffort"
EFFORT=high

The first one is the Default, which means Opus 5.5. No error, no warning. My high still works on Opus 5 and simply doesn't exist on Opus 5.5.

The fix is one more entry under modelSettings:

"claude-opus-5-5": { "effortLevel": "high" }

With it, the same probe returns EFFORT=high for default, for opus, and for the full name. You can also pick the level with /effort while on Opus 5.5, or from the /model picker: per the docs, that sets its level the same way the manual entry does.

Do you need to raise it? Not necessarily. Anthropic set it to medium, and at the same level Opus 5.5 thinks more per turn than Opus 5, especially at xhigh and max. Its medium isn't Opus 5's medium. What matters is that it's your call: if you set high for a reason, that reason no longer reaches it.

What changes without you touching anything

• Thinking can't be turned off. alwaysThinkingEnabled and MAX_THINKING_TOKENS=0 do nothing on Opus 5.5; effort is the only control left. The other knobs that aren't wired up anymore are in Claude Code thinking.
• 1M context on every plan, Pro included, with no [1m] suffix and no usage credits.
• /fast moves to Opus 5.5, now the fast mode default, at $8/$40 per MTok (it was $10/$50 on Opus 5).

Reference

Official docs: Model configuration · Introducing Claude Opus 5.5 · What's new in Claude Opus 5.5

Coming from Opus 5? What its launch brought, and why effort did carry over back then, is in Opus 5 in Claude Code. And if you're never sure which model to reach for, my actual routing is in Sonnet 5, Opus 5 and Fable 5.

Requirements

• Claude Code v2.1.280 or later (claude update). Every output above comes from that version or from 2.1.278. The Pro Default change comes from the 2.1.280 changelog; I haven't checked it on a Pro account.