Anthropic releases Claude Fable 5.2 and Mythos 5.2 with a paid fast mode
The two 5.2 models share one underlying network. Anthropic reports faster output, a 2.5 times fast mode at double the price, and about 55 per cent fewer safeguard blocks than Fable 5.1.
Anthropic published Claude Fable 5.2 and Claude Mythos 5.2 on 9 November 2026. According to the announcement the two are one underlying model offered in two ways. Fable 5.2 carries tighter security and biology safeguards and is open to all users, while Mythos 5.2 goes only to organisations vetted through Anthropic’s verification programmes for security and life sciences research. Anthropic has not disclosed the parameter count, context length or training data, and the model is closed and API-only.
The headline changes are speed and friction. Anthropic says output is generated up to 30 per cent faster than Fable 5.1, and an optional fast mode raises token generation by up to 2.5 times at double the price: $16 per million input tokens and $80 per million output tokens, against $8 and $40 for the standard tier. Safeguard false positives are reported to be about 55 per cent lower than on Fable 5.1. Preserved thinking is now on for all accounts, and the anti-distillation protections are said to cut harvested data by 90 to 99 per cent.
On benchmarks, the vendor reports 49.1 per cent on FrontierCode 2.0 against 32.4 per cent for Fable 5.1, 69.9 per cent on FrontierSWE v2 and 89.3 per cent on OSWorld 2.1. The model ID is claude-fable-5-2, available on the Claude Platform, AWS, Google Cloud and Microsoft Azure from launch. External testing included the UK AI Security Institute and METR.
Why it matters
For teams already on Fable 5.1 the upgrade is a model-ID change, but fast mode is a separate billing tier, so make sure your cost controls can tell the two apart before enabling it. The benchmark gains are Anthropic’s own numbers run at specific effort levels (the FrontierCode figure uses Xhigh effort, and High effort gives 45.3 per cent), so compare like with like when you rerun your own evaluations. The split between Fable and Mythos is a policy boundary rather than a capability one, which makes the verification programmes the thing to watch if your work touches security tooling or biology.