Anthropic announced the release of Claude Fable 5.2 and its restricted sibling Mythos 5.2 on Monday, positioning the new models as capable of turning a handful of instructions into complete software, physical designs, and research solutions. The company highlighted a 30% speed improvement over previous generations and a reduction in per‑token pricing: $8 per million input tokens and $40 per million output tokens, down from $10 and $50. Fast mode is available at double the rate, but the same speed advantage applies. According to the launch notes, Fable is intended for “serious work” that requires deep pockets, while Mythos is limited to verified organizations working in sensitive domains.

Cost of Large‑Scale Projects

AI YouTuber Matthew Berman tested the new models on three ambitious projects. A Gears of War‑style level required 8.63 billion tokens and cost an estimated $14,710 in API usage. An Excel clone consumed 5.5 billion tokens, translating to roughly $11,700 after discounts for cached inputs. A Fall Guys‑style game used 9 billion tokens over 71 hours, with an estimated cost of $13,000. Berman noted that the projects were not flawless, citing awkward animations and bugs, but he described Fable 5.2 as the “best AI model” he had used.

Pricing Strategy and Availability

The new token rates are still about twice the cost of Anthropic’s Claude Opus 5.5, which the company recommends for everyday tasks. For the first week after launch, Pro users and standard Team seats receive temporary Fable access or usage resets, capped at half their weekly allowance. Max, Ultra, and eligible premium seats get one to three resets, with resets for Max and Ultra lasting until March 8, 2027. The promotional access ends November 16, after which Fable returns to paid credits on all plans. Anthropic also announced that Mythos 5.2 will be available only to verified organizations.

Performance Benchmarks

Anthropic’s internal FrontierCode 2.0 evaluation shows Fable 5.2 achieving a 49.1% acceptance rate for proposed coding changes, up from 32.4% for Fable 5.1. The company claims the model solved two open theoretical‑physics problems involving black holes and designed three functioning electronic devices, though independent verification is pending. Crypto Briefing reported that Fable 5.2 set a record of 174 on the Epoch Capabilities Index, and that Peter Gostev compared it to Opus across more than 100 generations.

What to Watch Next

Analysts will be monitoring how developers balance the high cost of large projects against the speed and autonomy gains of Fable 5.2. The impact of the new fast mode on token consumption and pricing will also be a key factor. Finally, the rollout of Mythos 5.2 to restricted organizations may influence how sensitive industries adopt Anthropic’s technology.