Claude Fable 5.2 came out on 9 November 2026 and the first games made with it look a step above what people were making before. But it's not Anthropic's default recommendation for coding. That's still Claude Opus 5.5, at half the price. So which should you use to make a game?

Short answer: prototype and iterate with Opus 5.5; switch to Fable 5.2 for the long, unattended build where you want a finished, polished game at the end. The rest of this article explains why.

Price

These are Anthropic's API prices per million tokens:

Model Input Output Cache reads Fast mode (in / out)
Claude Fable 5.2 $8 $40 $0.20 $16 / $80
Claude Opus 5.5 $4 $20 $0.20 $8 / $40
Claude Sonnet 5.5 $2 $10 $0.10 —
Claude Fable 5.1 (previous) $10 $50 $0.25 —

Fable 5.2 is exactly twice Opus 5.5 on input and output, and 20% cheaper than Fable 5.1 was. Cache reads cost the same on both, which matters for games: long agent runs read their own project over and over, and most of that is cached. Saltline's 8-hour run used 445 million input tokens, mostly cached, and 2.7 million output tokens, for about $550.

On a Claude subscription you don't pay per token. Fable 5.2 can use up to 50% of your weekly usage limits on Pro plans and standard Team seats.

What Anthropic says each model is for

Anthropic is unusually direct about this. Its launch post says Fable and Mythos are "best for the most ambitious, open-ended technical work", and recommends starting with Opus 5.5 for most coding and knowledge tasks, then moving up to Fable or down to Sonnet or Haiku. The API docs put it the same way: start with Opus, and move to Fable 5.2 when Opus at higher effort still doesn't meet your needs.

Benchmarks that matter for game code

From Anthropic's launch numbers (their own evaluations, not independent tests):

  • FrontierCode 2.0: Fable 5.2 scores 49.1% at xhigh effort, the highest of any released model Anthropic tested. Opus 5.5's best is 45.5%.
  • The same benchmark at high effort: Fable 5.2 scores about 45.3%, level with Opus 5.5's best, at roughly half the cost per task.
  • FrontierSWE v2: 69.9% for Fable 5.2 vs 62.3% for Opus 5.5.

That second point is the interesting one. Fable 5.2 at high matches Opus 5.5 at its best, and costs less per task even though each token costs twice as much, because it gets there with fewer tokens. On hard tasks, the per-token price is not the whole story.

Anthropic also quotes Unity, which said Fable 5.2 was better at "building higher-quality 3D scenes from scratch" and "harder game-dev tasks".

What the launch-week games tell us

The Fable 5.2 games people shared in the first two days share a profile:

  • Saltline: one prompt, 8 hours, every model and sound generated in code.
  • Aiden's Revenge: one overnight prompt in Claude Code; the agent rewrote the game, generated art and audio, and wrote a bot to play all twelve stages.
  • The Clockmaker's Heart and Tidewright: first drafts with no human feedback.

These are all long, unattended runs that ended with a finished game. That's where Fable 5.2 earns its price. Its launch post highlights the same strength in other fields: a thousand agents formalizing a proof in 30 hours, an operating-system prototype built over eight days.

What they don't show is Fable 5.2 being much better at a quick edit. Fixing a jump that feels floaty or adding a pause menu is a short task. Opus 5.5, or Sonnet 5.5, will do it for less.

Which to use, by task

Task Use
Brainstorming a game idea, writing a design doc Opus 5.5 or Sonnet 5.5
First playable prototype you'll iterate on by hand Opus 5.5
Quick fixes: controls, UI, balancing numbers Opus 5.5 or Sonnet 5.5
An overnight, one-prompt build of a complete game Fable 5.2 at high
Procedural 3D assets, lighting and shaders from scratch Fable 5.2
A big refactor or porting the game to a new engine Fable 5.2
Many small parallel jobs (asset variations, level data) Sonnet 5.5 or Haiku 5.5 as subagents, Fable 5.2 as orchestrator

That last row is a pattern Anthropic mentions directly: Fable 5.2 works well as an orchestrator that delegates simpler tasks to smaller models, which keeps the bill down on big projects.

Speed

Fable 5.2's outputs are up to 30% faster than Fable 5.1's, and fast mode makes generation up to 2.5 times faster at double the price ($16 / $80). For a game you're building overnight, speed matters less than cost. For an interactive session where you're watching it work, fast mode on Opus 5.5 ($8 / $40, the same price as standard Fable 5.2) is another option.

Bottom line

  • On a budget, or iterating by hand: Opus 5.5.
  • Want a complete, polished game from one long run: Fable 5.2 at high effort, in an agent like Claude Code, with a prompt that asks it to play-test its own game. Our Three.js game guide has a prompt to start from.

To see what Fable 5.2 has made so far, browse our list of playable Claude Fable 5.2 games.