AGI-Eval Community Base - Subject Knowledge: leaderboard

AGI-Eval Community private-set Base Language Models evaluation (Subject Knowledge) measuring accuracy (%) across bilingual capability dimensions.

Metric: Accuracy (%). Source: agi-eval.org. Saturation forecast: Estimated already saturated. 25 models tracked. Updated 22 Nov 2026.

Top models

#ModelScoreOverall rank
1Kimi K2 Base72.57#939
2DeepSeek-V3.1-Base69.04
3ERNIE-4.5-300B-A47B-Base67.67
4DeepSeek V3 Base66.56
5DeepSeek-V3.2-Exp-Base66.29
6GLM-4.5-Base65.93
7Llama 4 Maverick Base61.44
8Qwen2.5-72B-Base59.43
9Qwen 2.5 32B (Base)58.11
10Qwen3-30B-A3B-Base57.03#1475
11Qwen3-14B-Base54.67#1473
12Qwen2.5-14B-Base52.89
13Qwen3-8B-Base51.12#1754
14Qwen3-4B-Base49.91#2439
15Llama 3.1 405B Base48.59

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

Interactive version: theaggregate.ai/benchmark?slug=agi-eval-community-base-subject-knowledge · How It Works · Data refreshed daily, snapshot 2026-11-22.