CAIS releases EnigmaEval, with Fable 5 leading and GPT-4o near zero
scaling01 · x · 2026-07-24
CAIS says EnigmaEval is now publicly available. It is a benchmark of long, complex reasoning problems that can take groups of people many hours or even days to solve.
The shared results show a wide spread across frontier models:
- Fable 5 leads the chart at 41.3 overall.
- GPT-5.6 Sol follows at 39.2.
- Gemini 3.1 Pro scores 32.4.
- Kimi K3 gets 23.6.
- Grok 4.5 reaches 17.0.
- Muse Spark 1.1 lands at 16.6.
- GPT-4o is far behind at 0.8.
A second chart breaks accuracy into normal puzzles and hard puzzles. On the hard set — described as problems that take groups of experts such as MIT students a few days — Fable 5 gets 10%, while GPT-5.6 Sol gets 5.1%.
More from Research
- Cognition's SWE-2 uses a KKT duality argument in RL to shift the effort Pareto curve — YouJiacheng · 2026-09-11
- VidMap uses RoMa coarse matching on all frames, fine-scale only for keyframes — ducha_aiki · 2026-09-11
- Bug Hunt Bench author: leaderboard noise is about 2-3 points — PawelHuryn · 2026-09-11
- PNAS paper shows a tiny billiard-ball system is a universal computer — undecidability lives in two dimensions — eigensteve · 2026-09-11
- New paper: Absolute pose estimation from affine cues and gravity direction — ducha_aiki · 2026-09-11
- LoMa Paper Ships REALLY HardPairs Dataset, Accepted at ECCV 2026 — ducha_aiki · 2026-09-11