Summary
Matt Wolfe takes a first look at Anthropic’s newly launched Claude Fable 5.2 and Mythos 5.2. The two are essentially the same model, with Mythos 5.2 as the top-end version limited mostly to cybersecurity testers and Fable 5.2 as the publicly available variant with additional cybersecurity guardrails. Anthropic recommends most users stick with Opus 5.5 as a daily driver and use Fable as an adviser or orchestrator that delegates simpler work to smaller models.
The video reviews the benchmark results, noting that gains in agentic coding are modest, such as 45.5% to 49.1% on Frontier Code, while agentic scientific research jumps to 74.6% from 58.7% for Opus 5.5. Google’s Gemini Carbon edges it out on business workflows. Wolfe also highlights a chart showing diminishing returns from extended thinking, where Fable 5.2 scores better at extra high effort than at max.
In hands-on testing, he asks the model to clone the game Mega Bonk. It declines to copy the title directly but builds an original game called Shardstorm with working mechanics, though he finds the graphics less polished than Opus 5.5’s. He also reviews Anthropic’s hardware design demos covering circuit boards, 3D printed enclosures and firmware.
📺 Source: Matt Wolfe · Published November 09, 2026
🏷️ Format: Review
