Claude Fable 5.2: $8/$40 per MTok, a fast mode, a new Claude Code default, and what one reviewer found, checked against Anthropic's pages

What the video shows

• Benchmark gains, as quoted from Anthropic's launch page. Opus 5.5 against Fable 5.2: agentic coding on Frontier Code 2.0, 45.5 against 49.1; agentic coding on STB V2, 62.3 against 69.9; agentic scientific research and business workflows, a 15 to 20 percent relative gain. His summary: about 5 to 15 percent better than Opus 5.5 across the board. These are vendor-published figures and were not independently run.
• A usage reset. Starting a new chat with Fable 5.2 and asking which model it is, he says, shows a notification in the app that gives a free usage reset. This is his description of what he saw, not a documented feature.
• Demos. He asked it for impressive demos he could show in under 10 minutes and it produced a typing demo, an SVG bicycle landing page, GPU reaction-diffusion simulations, cloth and fireworks simulations and a generated-landscape flyover. He puts the cost of the whole set at roughly 30 to 50 cents in tokens and says the work was one-shot with no optimisation.
• Mythos. A second model, Mythos, launched the same day but is available only to verified cybersecurity organisations.

Checked against Anthropic's pages

From the models overview and API release notes, and the Claude Code 2.1.321 release notes on GitHub, read 2026-11-10.

• ID and price. claude-fable-5-2, $8 input and $40 output per million tokens (Fable 5.1 was $10 and $50). Cache reads are 2.5% of the input price ($0.20 per million). 1M-token context window, 128K maximum output, always-on adaptive thinking, default effort high, retirement not sooner than 2027-11-09.
• Fast mode. Up to 2.5x higher output tokens per second at $16 input and $80 output per million tokens. Set speed: "fast" on the request; no beta header is needed.
• Data retention. Fable 5.2 and Mythos 5.2 require 30-day data retention and are not available under zero data retention unless Anthropic expressly authorises it.
• Platforms. The Claude API, Claude Platform on AWS and Claude Models on Google Cloud (Anthropic managed), plus Amazon Bedrock, Google-managed Claude Models on Google Cloud and Microsoft Foundry.
• Mythos 5.2. claude-mythos-5-2, limited to organisations verified through Anthropic's verification programs.
• Claude Code. 2.1.321 makes Fable 5.2 the default Fable model and moves saved Fable 5.1 picks to the fable alias: at once on Pro and Max plans, and on Team and Enterprise once the organisation allows it. /model claude-fable-5-1 switches back.

If you build on Fable

1. Pin claude-fable-5-2 explicitly in anything that must not change behaviour, rather than relying on the Claude Code fable alias.
2. Re-run your own evals before switching. The price fell by 20 percent but the benchmark gain is the vendor's, measured on the vendor's tasks.
3. Check data retention before moving a zero-data-retention workload.
4. If you used fast mode on Opus 4.8, note that it is removed on 2026-11-30 and requests then return an error; Anthropic points to Opus 5.5 for the migration.
5. Check your request shapes: per the API notes, edited-history requests that send back a thinking block return a 400 for every account, and the earlier computer_20251124 tool returns a 400 on every platform except Amazon Bedrock.

Caveats

• Most of the video is demonstration. Nothing in it shows a setting, command or prompt you could reuse, which is why the usable detail above is Anthropic's, not the creator's.
• The benchmark names and scores were read out from the launch page; check them there before quoting.
• "Mid-level or higher colleague" is the reviewer's opinion, not a measured result.