INSIGHT AI

Edition 2026-11-21 · retrieved 2026-11-20 09:18 UTC

Model Leaderboards

The same scores, filtered three ways: every model, models with published weights, and the best published-weights model you can run yourself at each memory size.

Model Leaderboard

Top models by Artificial Analysis Intelligence Index

#ChangeModelMakerScore
1–Claude Fable 5.2 (Max, Default Fallback)Anthropic64.6
2–GPT-6.2 Astra (Max)OpenAI62.1
3–Gemini 4 Neon (Xhigh)Google61.4
4–Gemini 4 Carbon (High)Google60.0
5–Claude Opus 5.5 (Max, Default Fallback)Anthropic59.1
6–GPT-6.2 Sol (Max)OpenAI57.3
7–GPT-6 Astra (Max)OpenAI56.9
8–GPT-6.1 Sol (Max)OpenAI54.6
9–Claude Sonnet 5.5 (Max, Default Fallback)Anthropic54.5
10–Gemini 4 Hydrogen (Xhigh)Google53.9

Open-Weights Leaderboard

Models whose weights the maker publishes on Hugging Face

#ChangeModelMakerScoreTotal paramsArchitectureMemory needLicenseWeights
1–DeepSeek V4.1 Flash 1017 (Max)DeepSeek32.6763B~457.9 GBmitdeepseek-ai/DeepSeek-V4.1-Flash-1017
2–GLM-5.3 (Max)Z AI31.7753B~452 GBotherzai-org/GLM-5.3
3–Ling 3.1 Flash 1019InclusionAI29.6572B~343.3 GBmitinclusionAI/Ling-3.1-flash-1019
4–Step 5 PreviewStepFun28.5604B~362.6 GBotherstepfun-ai/Step-5-Preview
5–GLM-5.3-FlashZ AI27.8321B~192.8 GBmitzai-org/GLM-5.3-Flash
6–Muse Spark Open (Xhigh)Meta24.9822BMoE, 42B active~493.2 GBothermeta-models/Muse-Spark-Open-816B-A42B
7–DeepSeek V4 Pro 0813 (Max)DeepSeek23.11.65T~990.3 GBmitdeepseek-ai/DeepSeek-V4-Pro-0813
8–Qwen3.8 27B (Xhigh)Alibaba20.127.8B~16.7 GBapache-2.0Qwen/Qwen3.8-27B
9–MiniMax-M3MiniMax18.9427B~256.2 GBotherMiniMaxAI/MiniMax-M3
10–K2 Horizon 375B A23BInstitute of Foundation Models17.8379BMoE, 23B active~227.5 GBapache-2.0IFM/K2-Horizon-375B-A23B

Best Model for 128 GB

Open-weights models whose 4-bit weights fit in 128 GB of unified memory or GPU VRAM

#ChangeModelMakerScoreTotal paramsArchitectureMemory needLicenseWeights
1–Qwen3.8 27B (Xhigh)Alibaba20.127.8B~16.7 GBapache-2.0Qwen/Qwen3.8-27B
2–Nemotron 3.5 Super VLNVIDIA13.6124BMoE, 12B active~74.7 GBothernvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16
3–Muse Glimmer (High)Meta6.329.8B~17.9 GBapache-2.0meta-models/Muse-Glimmer-30B
4–Nemotron 3.5 LightningNVIDIA5.531.6BMoE, 3B active~18.9 GBothernvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16
5–Granite 4.2 3BIBM4.73.7B~2.2 GBapache-2.0ibm-granite/granite-4.2-3b
6–Mistral Small 4 (Reasoning)Mistral4.3119B~71.6 GBapache-2.0mistralai/Mistral-Small-4-119B-2603
7–NVIDIA Nemotron 3 Nano 4BNVIDIA1.84.0B~2.4 GBothernvidia/NVIDIA-Nemotron-3-Nano-4B-BF16

Best Model for 64 GB

Open-weights models whose 4-bit weights fit in 64 GB of unified memory or GPU VRAM

#ChangeModelMakerScoreTotal paramsArchitectureMemory needLicenseWeights
1–Qwen3.8 27B (Xhigh)Alibaba20.127.8B~16.7 GBapache-2.0Qwen/Qwen3.8-27B
2–Muse Glimmer (High)Meta6.329.8B~17.9 GBapache-2.0meta-models/Muse-Glimmer-30B
3–Nemotron 3.5 LightningNVIDIA5.531.6BMoE, 3B active~18.9 GBothernvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16
4–Granite 4.2 3BIBM4.73.7B~2.2 GBapache-2.0ibm-granite/granite-4.2-3b
5–NVIDIA Nemotron 3 Nano 4BNVIDIA1.84.0B~2.4 GBothernvidia/NVIDIA-Nemotron-3-Nano-4B-BF16

Best Model for 32 GB

Open-weights models whose 4-bit weights fit in 32 GB of unified memory or GPU VRAM

#ChangeModelMakerScoreTotal paramsArchitectureMemory needLicenseWeights
1–Qwen3.8 27B (Xhigh)Alibaba20.127.8B~16.7 GBapache-2.0Qwen/Qwen3.8-27B
2–Muse Glimmer (High)Meta6.329.8B~17.9 GBapache-2.0meta-models/Muse-Glimmer-30B
3–Nemotron 3.5 LightningNVIDIA5.531.6BMoE, 3B active~18.9 GBothernvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16
4–Granite 4.2 3BIBM4.73.7B~2.2 GBapache-2.0ibm-granite/granite-4.2-3b
5–NVIDIA Nemotron 3 Nano 4BNVIDIA1.84.0B~2.4 GBothernvidia/NVIDIA-Nemotron-3-Nano-4B-BF16

Best Model for 16 GB

Open-weights models whose 4-bit weights fit in 16 GB of unified memory or GPU VRAM

#ChangeModelMakerScoreTotal paramsArchitectureMemory needLicenseWeights
1–Granite 4.2 3BIBM4.73.7B~2.2 GBapache-2.0ibm-granite/granite-4.2-3b
2–NVIDIA Nemotron 3 Nano 4BNVIDIA1.84.0B~2.4 GBothernvidia/NVIDIA-Nemotron-3-Nano-4B-BF16

Scores are the Artificial Analysis Intelligence Index. Each model appears once, at its best-scoring reasoning setting. A model counts as open weights only when its maker publishes the weights on Hugging Face; total parameters and license are read from the Hugging Face API.

Memory need = total parameters × 4-bit weights × 1.2 for KV cache and runtime. A model is on a memory board when that need fits in 75% of the memory, the share macOS gives the GPU by default. On Apple Silicon and other unified-memory machines (such as AMD Strix Halo) that is total RAM; on a discrete GPU it is VRAM. Mixture-of-experts models count every expert, because all of them must be loaded; their active parameters only change speed. On a discrete GPU, experts can be offloaded to system RAM to run larger models more slowly; these boards do not assume that. Changes compare with the previous edition.