Summary
Theo of t3.gg digs into the newly released Fable and Mythos 5.2 models from Anthropic, following the Opus 5.5, Sonnet 5.5, and Haiku 5.5 releases. He covers the headline results, including a state-of-the-art Artificial Analysis score reported roughly four and a half to five points ahead of the previous leader, Gemini 4 Carbon, along with claims about CAD work, formal proofs, and complex software migrations. He also explains that Fable and Mythos are the same underlying model.
The bulk of the video is hands-on testing. Theo runs the model against his own large agent-built codebases, including a TypeScript-in-Rust rewrite and T3 Code, and compares codebase analysis accuracy, cost, and speed with other models. Fable 5.2 made four false claims out of 61 in one analysis for about $23, while a cheaper rival made 38 claims with none false. In code review, it proved thorough and less prone to nitpicking but slow, taking about two hours for eight pull requests.
He also looks at real-world pricing, how quickly the model consumes subscription limits, and usage-share data from T3 Code and Slopolytics showing Claude’s share of traffic climbing to about 78% after Opus 5.5. The video closes with advice on where the model is worth using and where it frustrates.
📺 Source: Theo – t3․gg · Published November 10, 2026
🏷️ Format: Review
