Flame Machine Learning - Create: leaderboard

Metric: Accuracy (%; the paper's Bloom-level weighted accuracy on the Create problems of the 970-problem textbook-grounded Machine Learning multiple-choice set, zero-shot, temperature 0). Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)99
2GPT-5.491
3Claude Opus 4.688
4Gemini 2.5 Flash Lite83
5GPT-4.1 Mini82
6Qwen 3 32B75
7DeepSeek R1 Distill Qwen 32B66
8Claude Haiku 4.564
9DeepSeek R1 Distill Qwen 14B55
10Qwen 3 8B47
11Gemma 3 27B (IT)38
12Gemma 3 12B (IT)37

Interactive version: theaggregate.ai/benchmark?slug=flame-machine-learning-create · How It Works · Data refreshed daily, snapshot 2026-09-26.