Skip to content
Anish Gandhi Claude Opus 5.5 Connect on LinkedIn

← All guides · checked against Anthropic's docs on

Opus 5.5 in Claude Code

Claude Code handles the API changes for you. What is left is a new default, an effort setting that starts over, and a few commands that behave differently.

In 30 seconds

  • Set effort again. Opus 5.5 starts at medium. The level you saved before stays with your old model. Run /effort once on Opus 5.5.
  • Thinking stays on. The thinking toggle has no effect on Opus 5.5. Effort decides how much it thinks.
  • Another model may answer. Flagged security work re-runs on Opus 4.8, flagged biology on Opus 5, with a notice.
  • Cheaper. $4 / $20 per on an API key, and /fast at $8 / $40.

What changes for you

Open How to for where to click. Pick the model you used before to hide what does not apply to you.

Before Opus 5.5, I used

  1. Changes quietly

    Your saved effort level does not carry over

    Opus 5.5 starts at medium . A level you saved earlier with /effort keeps applying to your old model, not to Opus 5.5, until you pick one for it.

    • Coming from Opus 4.8 Opus 4.8 started at high.
    • Coming from Opus 5 Opus 5 started at high. At the same level, Opus 5.5 thinks more per turn, so medium is a fair place to start.
    • Coming from Fable 5.1 Fable 5.1 started at high.
    How to
    1. Switch to Opus 5.5, run /effort and choose a level. Claude Code saves it for Opus 5.5 only.
    2. Or set it per model under modelSettings in your . A top-level effortLevel in your user settings does not reach Opus 5.5. One in project, local or managed settings does.
    3. For one hard turn, write in the prompt instead of raising effort for the whole session.

    Source: Claude Code: model configuration

  2. Changes quietly If you used Opus 4.8 and Opus 5

    The thinking switch does nothing

    Opus 5.5 always . The Option+T toggle (Alt+T on Windows and Linux), alwaysThinkingEnabled and MAX_THINKING_TOKENS=0 have no effect on it. Effort decides how much it thinks.

    How to
    1. If you turned thinking off to save time or usage, lower the effort level instead.

    Source: Claude Code: model configuration

  3. Changes quietly

    Flagged requests re-run on another model

    A request flagged as security work re-runs on Opus 4.8, and one flagged as biology re-runs on Opus 5. A notice in the transcript says so.

    • Coming from Opus 4.8 Opus 4.8 has no checks like this. It is the model that security requests re-run on.
    • Coming from Opus 5 On Opus 5, security requests went to Opus 4.8 too, but biology requests ended in a refusal. Now they re-run on Opus 5.
    • Coming from Fable 5.1 Same routing as Fable 5.1.
    How to
    1. To decide each time: run /config and turn off , or set "switchModelsOnFlag": false. A flagged request then pauses with two choices: switch, or edit the prompt and retry.
    2. With it off, a non-interactive run (claude -p) ends in a refusal instead of pausing.
    3. --fallback-model is a different setting. It covers overloaded servers, not flagged requests.

    Source: Claude Code: model configuration

  4. Good to know

    Opus 5.5 becomes your default

    From Claude Code 2.1.280, the default model and the opus alias both mean Opus 5.5 on Pro, Max, Team, Enterprise and the API. On Pro and Team Standard, the default moved up from Sonnet.

    How to
    1. Update Claude Code, then run /model to see which model you are on.
    2. To stay on an older model, pick it in /model or set "model" in your .
    3. On Amazon Bedrock or Google Cloud, set ANTHROPIC_DEFAULT_OPUS_MODEL so a new Opus arrives when you choose, not when Claude Code updates.

    Source: Claude Code changelog, v2.1.280

  6. Good news

    Fast mode costs less, or exists at all

    /fast runs Opus 5.5 up to 2.5 times faster, at $8 / $40 per .

    • Coming from Opus 4.8 On Opus 4.8 it cost $10 / $50.
    • Coming from Opus 5 On Opus 5 it cost $10 / $50.
    • Coming from Fable 5.1 Fable 5.1 has no fast mode.
    How to
    1. On a subscription, fast mode is paid from , never from your plan limits. On Team and Enterprise an Owner has to turn it on.
    2. Turning it on partway through a session bills the whole conversation once at the full input price. Turn it on at the start.
    3. Not available on Amazon Bedrock, Google Cloud or Microsoft Foundry.

    Source: Claude Code: fast mode

  7. Good news

    Cheaper per token, and more room on plans

    On a Claude plan, your usage is shared across claude.ai, Claude Code and Desktop, and Anthropic raised the five-hour limits at launch. With an API key, you pay $4 / $20 per million tokens.

    • Coming from Opus 4.8 API-key price on Opus 4.8: $5 / $25.
    • Coming from Opus 5 API-key price on Opus 5: $5 / $25.
    • Coming from Fable 5.1 API-key price on Fable 5.1: $10 / $50.
    How to
    1. /usage shows what a session has used, or its estimated cost on an API key.
    2. The saved is not available in the terminal. Use it from Settings > Usage in a browser or Claude Desktop.

    Source: Claude Code: costs

What stays the same

  • Claude Code already keeps the way Opus 5.5 requires. Nothing to change.
  • Refused and flagged requests are handled for you. The setting above is the only control.
  • A 1M-token on every plan, including Pro, with no extra charge. Sessions on their own near 967K tokens.
  • ultrathink still asks for deeper thinking on one turn.

Sources

Every fact on this page comes from Anthropic's own documentation, checked on 23 Sep 2026, unless it is labelled as a tester's report. Numbers from testers come from one small team with early access. Treat them as early impressions.

Anthropic's documentation