Fable 5.2 Leads MazeBench Token Efficiency; Billion-Token Runs Underway

patience_cave · x · 2026-11-16

MazeBench's author reports Fable 5.2 is the most token-efficient model ever tested, scoring far above all other agents at the 200M-token checkpoint (21% at 179.8M tokens vs GPT-6 Astra's 16%, Gemini 4 Argon 9%, Opus 5.5 7%).

Since frontier models keep improving past 200M tokens, each model will be run to one billion tokens; Astra's run is complete and Fable's is still in progress.

Related event: Fable 5.2 Leads MazeBench Token Efficiency; Billion-Token Runs Underway(2 posts)→

Original post →

More from Models

Models channel →