LiveMathBench 202412 - Hard: leaderboard

The 45 LiveMathBench 202412 questions (21 English, 24 Chinese) on which QwQ-32B-Preview and o1-mini did worst, 48 sampled answers each; the score is mG-Pass@16. The two selecting models and the answer judge (Qwen2.5-72B-Instruct) are withheld from this board.

Metric: mG-Pass@16 on the 45-question hard split (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 18 models tracked. Updated 17 Nov 2026.

Top models

#ModelScoreOverall rank
1DeepSeek R129.6#487
2O3 Mini28.6#547
3DeepSeek R1 Distill Llama 70B17.3#1669
4DeepSeek R1 Distill Qwen 14B15.5#2683
5DeepSeek R1 Distill Qwen 32B15.1#1806
6Qwen 2.5 32B Instruct9.1#1257
7DeepSeek-R1-Distill-Qwen-7B8.1#7957
8Qwen2.5-Math-72B-Instruct7.9#1757
9NuminaMATH-72B-COT7.3
10InternLM3-8B-Instruct7.1#2679
11Llama 3.1 70B Instruct7#1577
12DeepSeek R1 Distill Llama 8B6.2#6937
13DeepSeek R1 Distill Qwen 1.5B4.6#9148
14Llama 3.3 70B Instruct4.4#1387
15Qwen2.5-Math-7B-Instruct3.8#8400

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=livemathbench-202412-hard · How It Works · Data refreshed daily, snapshot 2026-11-17.