Wes Roth · 2026-11-10 · notable
Wes Roth — 'NOTHING will be smarter than Fable 5.2...'
In short
Wes Roth reviews Claude Fable 5.2 and Mythos 5.2, focusing on Anthropic's note that it limited Fable 5.2's training on kernel optimization and LLM architecture research, and tests Fable 5.2 on his own game benchmark.
Wes Roth · 2026-11-10 · 3 sources checked
Wes Roth reads the fine print of the Claude Fable 5.2 launch and tests the model live.
What is it?
'NOTHING will be smarter than Fable 5.2...' is a 16-minute Wes Roth video uploaded on 10 November 2026 about Anthropic's Claude Fable 5.2 and Mythos 5.2 launch. Roth argues the biggest story is in the fine print: Anthropic says it deliberately limited Fable 5.2's training on accelerator kernel optimization and LLM architecture research, keeping those skills at Opus 5.5 levels.
How does it work?
Chapters move from the black-hole math results and the training limits to METR's report, which Roth argues confirms Fable 5.2 is METR's 'Model 3', and to Anthropic's NanoGPT-style experiment scaffold. The video ends with a live test on a custom game benchmark and a look at usage limits. The description links the launch post, the 264-page system card and the METR assessment.
Why does it matter?
The training limits are easy to miss in a 264-page system card. The video is a short guided reading for viewers who want to know what Anthropic held back and why it matters for AI research speed-up.
Who is it for?
people following frontier models and AI safety