Matt Shumer's Fable 5.2 review: smartest model yet, but Max effort scores lower at $44.70/task
felixrieseberg · x · 2026-11-10
Matt Shumer's long-form review calls Fable 5.2 the most capable model he's used — trustworthy enough to run chunks of a company autonomously. Single-agent ownership works best: it built a Hogwarts-meets-Fortnite battle royale unaided, nailed a voxel engine with real destruction physics running on a MacBook, and shows unprecedented design skill (with many style examples). Downsides: overly cautious 'hall monitor' behavior, Claude-speak, and the official chart shows Fable 5.2 Max scores 45.6% on FrontierCode 2.0 at $44.70/task — worse and pricier than Xhigh (49.1%).
Related event: Shumer Reviews Fable 5.2: Smartest Model Ever, But Heavy Guardrails(3 posts)→
More from Models
- Claude Fable 5.2 designs 3 working electronic devices from one-paragraph prompts — Dr_Singularity · 2026-11-10
- Fable 5.2 costs $26.54 per task, pricier than Fable 5.1 despite cheaper claims — BLUECOW009 · 2026-11-10
- Salesforce's Interactive Visual Gym: best VLM finishes only 53.4% of 315 closed-loop tasks — jcniebles · 2026-11-10
- Claude Fable 5.2 launch: subscribers get 1-3 usage resets based on tenure — bcherny · 2026-11-10
- Benchmark audit: 5 of 6 AI benchmarks have grading defects; only PostTrainBench verified — maksym_andr · 2026-11-10
- Matt Shumer charts cursing at agents: 46 swears in 3 days on Fable 5.2 vs 38 in 26 days on Opus 5.5 — mattshumer_ · 2026-11-10