Daily Empirical IndexSnapshot 2026-11-0110 Open-Weight Models0.13M 30d HF Pulls3.7K Likes
Open Source Uncensored LLMs Leaderboard
Empirical rankings of open-weight uncensored and abliterated language models evaluated across UGI adversarial compliance, MMLU-Pro reasoning retention, EQ-Bench creative nuance, and quantized GPU VRAM sizing.
★ #1 Overall Best ModelGlobal #1 · Score 93.5
UGI 78.2·MMLU 76.5·96.8% Comp.
Coding & STEM
405B·228 GB Q4·MMLU 76.5
Creative & Roleplay
405B·228 GB Q4·EQ 88.7
Deep Reasoning
405B·228 GB Q4·UGI 78.2
Zero-Refusal
32.5B·20 GB Q4·98.5% Comp.
8 GB VRAM GPU
8B·5.5 GB Q4·128k ctx
16 GB VRAM GPU
24B·15 GB Q4·32k ctx
24 GB VRAM GPU
32.8B·20 GB Q4·128k ctx
48 GB VRAM GPU
32.8B·20 GB Q4·128k ctx
VRAM:
| Rank | Actions | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| #1 | Llama 3.1·405B·128k ctx·NousResearch | 93.5 | 78.2 | 96.8% | 76.5 | 88.7 | 228 GB | 9,684 | 476 | |
| #2 | Qwen 2.5·32.8B·128k ctx·huihui-ai | 90.8 | 74.9 | 97.4% | 74.2 | 81.8 | 20 GB | 18,500 | 460 | |
| #3 | Llama 3.3·70.6B·128k ctx·huihui-ai | 90.6 | 75.8 | 97.6% | 70.9 | 85.4 | 41 GB | 11,420 | 312 | |
| #4 | Qwen 2.5·72.7B·128k ctx·huihui-ai | 90.4 | 76.4 | 98.2% | 71.8 | 84.6 | 42 GB | 8,420 | 288 | |
| #5 | Qwen 2.5 Coder·32.5B·128k ctx·huihui-ai | 90.0 | 73.9 | 98.5% | 73.1 | 78.4 | 20 GB | 15,940 | 415 | |
| #6 | Mistral Small 3·24B·32k ctx·cognitivecomputations | 88.8 | 74.1 | 99.1% | 66.4 | 82.9 | 15 GB | 14,200 | 342 | |
| #7 | Qwen 2.5·72.7B·64k ctx·EVA-UNIT-01 | 88.5 | 73.5 | 98.9% | 68.1 | 89.4 | 42 GB | 4,102 | 215 | |
| #8 | Qwen 2.5·72.7B·32k ctx·anthracite-org | 88.2 | 72.8 | 99.4% | 66.8 | 87.9 | 42 GB | 6,136 | 264 | |
| #9 | Gemma 2·27.2B·8k ctx·mlabonne | 87.4 | 71.2 | 97.8% | 64.5 | 85.1 | 17 GB | 9,820 | 368 | |
| #10 | Llama 3.1·8B·128k ctx·Orenguteng | 84.6 | 67.4 | 99.2% | 56.8 | 79.2 | 5.5 GB | 28,632 | 538 |
Showing 1–10 of 10 models·Page 1 of 1