Qwen 3
Best conversion per model
One row per base model. The full list of attempts is on each model page.
| # | Model | Author | Params | KL | Refusals | Score | Evidence | Downloads |
|---|---|---|---|---|---|---|---|---|
| 1 | Qwen3-30B-A3B-Gemini-Pro-High-Reasoning-2507-HI8 DavidAU | ayunami2000 | 30B · 3B active | 0.0014 | 2/1002% | 98.9 | BestParetoAWQEXLGGUF | 11.9K |
| 2 | Qwen3-0.6B-Base Qwen | Iwaku-Real | 0.6B | 0.0027 | 2/1002% | 98.9 | BestParetoSAFETENSORS | 572 |
| 3 | Qwen3-VL-4B-Thinking Qwen | sh0ck0r | 4B | 0.0004 | 3/1003% | 98.5 | BestParetoSAFETENSORS | 80 |
| 4 | Qwen3-0.6B Qwen | Roman0 | 0.6B | 0.0024 | 3/1003% | 98.4 | BestParetoSAFETENSORS | 582 |
| 5 | Qwen3-VL-32B-Thinking Qwen | coder3101 | 32B | 0.0048 | 3/1003% | 98.3 | BestParetoGGUFSAFETENSORS | 47.1K |
| 6 | Qwen3-4B-Thinking-2507 Qwen | Daizee | 4B | 0.0089 | 3/1003% | 98.1 | BestParetoMLXSAFETENSORS | 1.2K |
| 7 | Qwen3-VL-8B-Thinking Qwen | ZuzeTt | 8B | 0.030 | 1/1001% | 98.0 | BestParetoFP8GGUFNVFP4 | 16.9K |
| 8 | Qwen3-4B-Thinking-2507 unsloth | JoaoZaokk | 4B | 0.0025 | 4/1004% | 97.9 | BestParetoAWQGGUFGPTQ | 555 |
| 9 | Qwen3-VL-2B-Thinking Qwen | sh0ck0r | 2B | 0.0053 | 4/1004% | 97.7 | BestParetoGGUFSAFETENSORS | 6.5K |
| 10 | Qwen3-1.7B Qwen | saidutta69 | 1.7B | 0.025 | 4/1004% | 96.8 | BestParetoGGUFSAFETENSORS | 3.4K |
| 11 | Qwen3-VL-4B-Instruct Qwen | viaqlo | 4B | 0.026 | 4/1004% | 96.7 | BestParetoSAFETENSORS | 125 |
| 12 | Qwen3Guard-Gen-0.6B Qwen | rainmana | 0.6B | 0.072 | 0/1000% | 96.6 | BestParetoGGUFSAFETENSORS | 6.9K |
| 13 | Qwen3-4B-Base Qwen | Lambent | 4B | 0.040 | 2/554% | 96.3 | BestParetoGGUFSAFETENSORS | 4.5K |
| 14 | Qwen3-4B-Instruct-2507 Qwen | ma2t4 | 4B | 0.028 | 5/1005% | 96.1 | BestParetoGGUFSAFETENSORS | 147 |
| 15 | Qwen3-VL-32B-Instruct Qwen | llmfan46 | 32B | 0.042 | 4/1004% | 96.0 | BestParetoAWQGGUFNVFP4 | 192.7K |
| 16 | Qwen3-VL-8B-Instruct Qwen | heretic-org | 8B | 0.021 | 6/1006% | 96.0 | BestGGUFSAFETENSORS | 17.1K |
| 17 | Qwen3-14B Qwen | avi686 | 14B | 0.050 | 5/1005% | 95.1 | BestGGUFSAFETENSORS | 1.4K |
| 18 | Qwen3-Swallow-30B-A3B-RL-v0.2 tokyotech-llm | sasa2000 | 30B · 3B active | 0.010 | 9/1009% | 95.0 | BestParetoAWQGGUFGPTQ | 12.7K |
| 19 | Qwen3-VL-2B-Instruct Qwen | saidutta69 | 2B | 0.053 | 5/1005% | 95.0 | BestParetoGGUFSAFETENSORS | 2K |
| 20 | Qwen3-4B-Instruct-2507 unsloth | 0xA50C1A1 | 4B | 0.066 | 6/1006% | 93.9 | BestParetoGGUFSAFETENSORS | 7.2K |
| 21 | Qwen3-Coder-30B-A3B-Instruct Qwen | jarradh | 30B · 3B active | 0.104 | 3/1003% | 93.8 | BestParetoGGUFLORASAFETENSORS | 4K |
| 22 | Qwen3-4B-Instruct-2507-Polaris-Alpha-Distill TeichAI | DavidAU | 4B | 0.060 | 8/1008% | 93.2 | BestParetoAWQEXLGGUF | 18K |
| 23 | Qwen3-Esper3-Reasoning-CODER-Instruct-12B-Brainstorm20x DavidAU | avi686 | 12B | 0.067 | 8/1008% | 92.9 | BestParetoAWQEXLGGUF | 298 |
| 24 | Qwen3-4B-Thinking-2507-Claude-4.5-Opus-High-Reasoning-Distill TeichAI | DavidAU | 4B | 0.010 | 14/10014% | 92.5 | BestParetoAWQEXLGGUF | 18.4K |
| 25 | typhoon2.5-qwen3-30b-a3b typhoon-ai | tgncr | 30B · 3B active | 0.127 | 4/1004% | 92.3 | BestParetoSAFETENSORS | 739 |
| 26 | Qwen3-30B-A3B-Instruct-2507-REAM SamsungSAILMontreal | sasa2000 | 30B · 3B active | 0.199 | 2/1002% | 90.7 | BestParetoGGUFSAFETENSORS | 22.4K |
| 27 | Qwen3-Coder-Next Qwen | trohrbaugh | 79.7B | 0.071 | 14/10014% | 89.7 | BestParetoGGUFSAFETENSORS | 59.4K |
| 28 | GPT-5-Distill-Qwen3-4B-Instruct Jackrong | ChiKoi7 | 4B | 0.268 | 10/10010% | 84.4 | BestParetoGGUFSAFETENSORS | 16.1K |
| 29 | typhoon2.5-qwen3-4b typhoon-ai | sthaps | 4B | 0.176 | 26/10026% | 79.5 | BestParetoSAFETENSORS | 50 |
All models
Unmeasured models
heretic_Qwen3-1.7B-MegaScience, Heretic-Qwen3-0.6B-Diagnose, Heretic-Qwen3-0.6B-Psychological-Support, Heretic-Qwen3-0.6B-Treatment, Heretic-Qwen3-MOE-4x0.6B-2.4B-Writing-Thunder, Heretic-Qwen3-Zero-Coder-Reasoning-V2-0.8B, muse-qwen3vl-heretic, Qwen3-0.6B-Heretic, Qwen3-0.6B-heretic-Base, Qwen3-0.6B-heretic-Base2, Qwen3-0.6B-heretic-OG, Qwen3-0.6B-heretic-Reproduce2, Qwen3-1.7B-Magic_heretic, Qwen3-14B-Heretic, Qwen3-14B-heretic-tools, qwen3-14b-heretic-uncensored, Qwen3-24B-A4B-Freedom-HQ-Thinking-Abliterated-Heretic, Qwen3-30B-A3B-Instruct-2507, Qwen3-32B, Qwen3-32B-VL-GLM-4.7-Flash-HI16-Heretic-Uncensored-Thinking, Qwen3-32B-VL-Instruct-Polaris-HI16-Heretic-Uncensored, Qwen3-4B, Qwen3-4B-Apollo-V0.1-Thinking-heretic-Uncensored-Abliterated, qwen3-4b-base-unsloth-bnb-4bit, Qwen3-4B-Claude-Sonnet-4-Reasoning-Distill-Safetensor, Qwen3-4B-Cybersecurity-GGUF, qwen3-4b-heretic-zimage, Qwen3-4B-Instruct-2507-abliterated-1128, Qwen3-4B-Instruct-2507-heretic, Qwen3-4B-Instruct-2507-heretic-av2, Qwen3-4B-Instruct-2507-heretic-v3-quantized-processing, Qwen3-4B-Jan-v1-Heretic-Abliterated, Qwen3-4B-RA-SFT, Qwen3-4B-Thinking-2507-Gemini-2.5-Flash-Lite-Preview-Distill, Qwen3-4B-Thinking-2507-Gemini-3-Pro-Preview-High-Reasoning-Distill-Heretic-Abliterated, Qwen3-4B-Thinking-2507-Gemini-3-Pro-Preview-High-Reasoning-Distill-Heretic-qx86-hi-mlx, Qwen3-4B-Thinking-2507-heretic, Qwen3-4B-Thinking-2507-Heretic, qwen3-4b-Z-Image-Engineer-V2-4bit-MLX, qwen3-4b-Z-Image-Engineer-V2-8bit-MLX, Qwen3-Omni-30B-A3B-Thinking, Qwen3-Swallow-8B-RL-v0.2, Qwen3-VL-235B-A22B-Instruct, Qwen3-VL-235B-A22B-Thinking-heretic, Qwen3-VL-30B-A3B-Instruct, Qwen3-VL-30B-A3B-Thinking, Qwen3-VL-32B-Heretic-v1, Qwen3-VL-32B-Heretic-v2, Qwen3-VL-32B-Instruct-Heretic-Text, Qwen3-VL-4B-Instruct-heretic-Semantic-Same, Qwen3-VL-4B-Instruct-Uncensored, Qwen3-VL-4B-Instruct-Uncensored-abliterated, Qwen3-VL-8B-Instruct-Unredacted-MAX, Qwen3-VL-8B-NSFW-Caption-V4.5-heretic, Qwen3Guard-Gen-4B, Qwen3Guard-Gen-8B-DeathGuard-heretic, raw-uncensored-qwen3-14b-heretic — conversions exist but their publishers did not report KL and refusal numbers, so they are not ranked.