Category ranking

Best AI models for knowledge

Knowledge tests cover broad academic recall, short factual answers where hallucination is penalised, image-based college questions, and how precisely a model follows explicit formatting instructions. Across 4 models with at least one knowledge score, Nemotron 3 Ultra has the highest index at 99.6. Among models you can deploy on OpenLLM Buddy, Qwen3.6 27B leads, ranked 2 overall.

Knowledge index

Average of each model's scores relative to the best published score on every test in this category (top = 100). Only tests a model reports are counted, and a model needs at least two of them to be ranked.

  • OpenLLM Buddy catalog
  • Open weights
  • Closed API
  1. 1Nemotron 3 UltraNVIDIA · 2 of 6 tests99.6
  2. 2Qwen3.6 27BAlibaba Qwen · 2 of 6 tests99.3
  3. 3Gemma 4 31BGoogle · 2 of 6 tests92.5
  4. 4Gemma 4 26B A4BGoogle · 2 of 6 tests89.3

10 more models report a single knowledge test and are listed in the table below without a rank: Kimi K3, Qwen3.8 27B, Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, Gemini 3.8 Flash, DeepSeek V4 Pro, Kimi K3.1 Pro, Step 5 Preview, Reflection Beam 1.0.

Raw scores by test

Published percentages. A dot means the model has no published score on that test.

ModelMMLU-ProSimpleQAIFEvalIFBenchMMMUMMMU-Pro
Nemotron 3 Ultra86.8··81.7··
Qwen3.6 27B86.2···82.9·
Gemma 4 31B85.2····76.9
Gemma 4 26B A4B82.6····73.8
Kimi K3·····81.6
Qwen3.8 27B···79.5··
Claude Opus 5.5·····87.7
GPT-6 Astra·····86.9
GPT-6.1 Sol·····86.0
Gemini 3.8 Flash·····85.6
DeepSeek V4 Pro87.5·····
Kimi K3.1 Pro·····84.1
Step 5 Preview·····76.0
Reflection Beam 1.0···79.6··
MMLU-ProBroad academic knowledge with ten answer options instead of four.5 models · top: DeepSeek V4 Pro 87.5
SimpleQAShort fact-seeking questions that expose hallucination.No published scores yet
IFEvalVerifiable formatting and content instructions.No published scores yet
IFBenchOut-of-distribution instruction-following constraints.3 models · top: Nemotron 3 Ultra 81.7
MMMUCollege-level questions that require reading images.1 models · top: Qwen3.6 27B 82.9
MMMU-ProA harder, filtered version of MMMU with ten options and vision-only questions.9 models · top: Claude Opus 5.5 87.7

Scores are published results from model cards, lab reports, and public leaderboards, each linked to its source. Settings such as thinking mode or test version differ between sources and are shown next to each score. OpenLLM Buddy catalog models are uncensored fine-tunes; their scores are those of the official base model. Data collected 2026-10-30. How the index works