vaultbox/ qwen3.5-uncensored:4b

12.6K 1 month ago

www.vaultbox.ai

vision tools thinking 4b 9b 27b 35b 
ollama run vaultbox/qwen3.5-uncensored:4b

Details

1 month ago

388a0b599356 · 3.4GB ·

model
qwen35
·
4.66B
·
Q4_K_M
params
{ "presence_penalty": 1.5, "temperature": 1, "top_k": 20, "top_p": 0.95 }
template
{{ .Prompt }}

Readme

www.vaultbox.ai

This model may produce uncensored content, including explicit, sensitive, or otherwise objectionable material. All outputs are generated automatically and are provided “as is,” without warranties of any kind.

The provider does not endorse or assume responsibility for any generated content. Users assume full responsibility for all use, distribution, and consequences arising from the use of this model.

This a comparison to the base ollama qwen models and the huihui_ai/qwen3.5-abliterated uncensored model on ollama.

Full Results Table — score (correct/total)(runs), score % higher is better

Model HumanEval TruthfulQA IFEval Uncensored↑ GSM8K-CoT
qwen3.5:4b 74.1 (243⁄328)(2) 46.2 (755⁄1634)(2) 81.0 (876⁄1082)(2) 2.8 (23⁄832)(2) 83.9 (2212⁄2638)(2)
vaultbox/qwen3.5-uncensored:4b 71.6 (235⁄328)(2) 42.8 (700⁄1634)(2) 80.2 (868⁄1082)(2) 97.7 (813⁄832)(2) 82.7 (2181⁄2638)(2)
huihui_ai/qwen3.5-abliterated:4b 57.3 (188⁄328)(2) 33.9 (554⁄1634)(2) 63.8 (690⁄1082)(2) 100.0 (832⁄832)(2) 75.0 (1978⁄2638)(2)
qwen3.5:9b 79.0 (259⁄328)(2) 51.8 (846⁄1634)(2) 81.4 (881⁄1082)(2) 2.2 (18⁄832)(2) 87.6 (2311⁄2638)(2)
vaultbox/qwen3.5-uncensored:9b 80.2 (263⁄328)(2) 46.0 (752⁄1634)(2) 79.4 (859⁄1082)(2) 99.5 (828⁄832)(2) 86.2 (2275⁄2638)(2)
huihui_ai/qwen3.5-abliterated:9b 72.0 (236⁄328)(2) 38.4 (628⁄1634)(2) 82.6 (894⁄1082)(2) 99.6 (829⁄832)(2) 82.6 (2180⁄2638)(2)
qwen3.5:27b 95.7 (314⁄328)(2) 55.0 (898⁄1634)(2) 87.5 (947⁄1082)(2) 0.6 (5⁄832)(2) 92.0 (2427⁄2638)(2)
vaultbox/qwen3.5-uncensored:27b 94.8 (311⁄328)(2) 47.9 (782⁄1634)(2) 86.1 (932⁄1082)(2) 98.3 (818⁄832)(2) 91.8 (2421⁄2638)(2)
huihui_ai/qwen3.5-abliterated:27b 93.9 (308⁄328)(2) 51.4 (840⁄1634)(2) 86.5 (936⁄1082)(2) 44.8 (373⁄832)(2) 92.3 (2434⁄2638)(2)
qwen3.5:35b 90.9 (298⁄328)(2) 51.8 (847⁄1634)(2) 85.2 (922⁄1082)(2) 0.5 (4⁄832)(2) 90.4 (2386⁄2638)(2)
vaultbox/qwen3.5-uncensored:35b 92.7 (304⁄328)(2) 46.3 (757⁄1634)(2) 86.4 (935⁄1082)(2) 100.0 (832⁄832)(2) 88.4 (2331⁄2638)(2)
huihui_ai/qwen3.5-abliterated:35b 89.0 (292⁄328)(2) 43.9 (718⁄1634)(2) 83.5 (903⁄1082)(2) 92.1 (766⁄832)(2) 88.6 (2336⁄2638)(2)

Delta Table: variant − base (pp, positive = variant wins)

Variant HumanEval TruthfulQA IFEval Uncensored↑ GSM8K-CoT
4b Δvaultbox -2.4 -3.4 -0.7 +95.0 -1.2
4b Δhuihui -16.8 -12.3 -17.2 +97.2 -8.9
9b Δvaultbox +1.2 -5.8 -2.0 +97.4 -1.4
9b Δhuihui -7.0 -13.3 +1.2 +97.5 -5.0
27b Δvaultbox -0.9 -7.1 -1.4 +97.7 -0.2
27b Δhuihui -1.8 -3.5 -1.0 +44.2 +0.3
35b Δvaultbox +1.8 -5.5 +1.2 +99.5 -2.1
35b Δhuihui -1.8 -7.9 -1.8 +91.6 -1.9

Apache 2.0 License