Claude Fable 5.2 is out: when to use it for code and games

What was released

On 9 November, Anthropic officially introduced Claude Fable 5.2 alongside Claude Mythos 5.2. They share an underlying model. Fable is broadly available with safeguards for sensitive work, while Mythos is restricted to vetted organizations in cybersecurity and life sciences. They are not two equally accessible choices in a chat menu.

Fable 5.2 accepts text and images and produces text. Its documentation lists a context window of up to one million tokens and a maximum output of 128,000 tokens. Context is the amount of material a request can include; it does not guarantee that the model will correctly understand every file in a large repository. Its API model ID is claude-fable-5-2.

Sources for this section: [1] · [2]

What changed, and how to read the numbers

In Anthropic's published tests, Fable 5.2 scores 49.1% on FrontierCode 2.0 at the highest effort setting versus 32.4% for Fable 5.1. On FrontierSWE v2 the company reports 69.9% versus 56.3%. These are results on particular task sets and settings, not the probability that the model will build your game without errors.

Anthropic also says output generation is up to 30% faster than version 5.1. An optional fast mode can raise output speed by up to 2.5 times, at a higher price. “Up to” matters: real timing depends on response length, load, tools, and the task.

Anthropic includes early feedback from a Unity specialist about better 3D scenes and harder game development tasks. That is one tester's experience, not our independent Maya or Unity test. The model does not automatically gain access to your scene: editing requires a suitable tool, an integration, and review of the result.

Sources for this section: [1] · [2]

What it costs

In the Claude API, Fable 5.2 costs $8 per million input tokens and $40 per million output tokens at standard rates. For comparison, Opus 5.5 is $4/$20 and Sonnet 5.5 is $2/$10. Tokens roughly correspond to pieces of words; input includes your request and supplied context, output is what the model generates.

For example, 100,000 input and 20,000 output tokens cost about $1.60 with Fable or $0.80 with Opus at base API rates. Fable fast mode is $16/$80, making that example about $3.20. This is API arithmetic only: caching, tools, other charges, and actual token counts can change the bill. A Claude web subscription has its own limits and is not a prepaid API token balance.

Sources for this section: [1] · [2]

Estimate the cost of your API request

Enter approximate input and output tokens and the number of identical runs. The calculator compares Fable 5.2, fast mode, Opus 5.5, and Sonnet 5.5 locally in your browser.

Sources for this section: [2]

When Fable is useful, and when Opus is enough

Start ordinary development with Opus 5.5. Anthropic itself recommends this. If Opus still struggles with a demanding task after clear requirements, access to the right files, and tests, try Fable 5.2. Do not select a model from a benchmark table alone: compare the corrected result, time, and cost on the same task.

• For copy, a button, simple layout, or one small bug: try Sonnet or Opus first. Fable may not justify the price difference.
• For a cross-module bug, a large migration, or a long project with many checks: Fable is worth testing.
• For a game with unstable logic, saves, mobile controls, and several scenes: give the model one bounded stage and verify it in the browser.
• For Maya or Unity, use Fable to analyze screenshots, write tool code, and plan a rig; scene access must be configured separately.

Sources for this section: [1] · [2]

How to try it without overspending

The launch offer applies only to eligible paid plans. Anthropic temporarily includes Fable in Pro and standard Team seat limits through 16 November 2026 at 11:59 p.m. Pacific, capped at 50% of weekly usage; afterward separate usage credits may be needed. Other eligible plans receive temporary limit resets. Check the terms and your balance under Settings → Usage before a long session.

1. Sign in to Claude at claude.ai. Open the model picker and check whether Fable 5.2 is available on your plan. The launch offer does not apply to the Free plan.
2. In Claude Code, open /model and select Fable 5.2 if it appears. For API use the model ID claude-fable-5-2 and check your account's spending limit.
3. Pick one real task from your game or app. Save the project state, give Opus and Fable the same requirements, then compare tests, defects, time, and tokens used.

Sources for this section: [3] · [4]

A practical test on your own game

Copy the task below and replace the bracketed parts. Give Opus and Fable identical copies of your project. This is an editorial test template, not a claim that we have already tested Fable on this game.

Work on my game [name/folder]. Problem: [reproducible bug]. Expected behavior: [what should happen]. First locate the files responsible for this flow and briefly explain the likely cause. Then make the smallest fix without rewriting the game. Do not change authentication, payments, or unrelated files. Add a check that would actually catch this bug; run available tests and the build. If you cannot run something, say exactly what and why. At the end, list changed files, check results, remaining uncertainty, and tokens/time used if available. Do not publish the game without my separate decision.

For 3D animation, replace the bug description with your rig issue and add a screenshot, Maya version, controller names, and the desired joint bend. Ask the model to separate what it saw in the image from what it inferred. Verify scene access and deformation quality in Maya itself.

Compare more than report quality: whether the bug is fixed, checks pass, neighboring features still work, and what it cost. If Opus does the job just as well, keep using it for that task type.

Sources for this section: [4]

What the release does not promise

A million-token window does not replace selecting the relevant files. A high benchmark score does not prove bug-free code in your project. Unity's feedback does not mean the model can rig or animate characters in Maya on its own. Faster output mode means faster text generation under certain conditions, not a finished game in half the time.

Before uploading work or other people's private data, check the retention terms: Fable 5.2 documentation describes 30-day retention by default and separate authorization for zero-data-retention access. Do not put credentials or personal data into a sample project.

Sources for this section: [2]