Most "best uncensored model" lists are vibes. This one is measured. We run every uncensored and abliterated model on the Open Uncensored API through the same tests and combine them into one number, the OpenUncensored Score: how smart the model it was made from is (40%), whether it finishes agent tasks without looping or crashing (30%), whether it actually answers edgy requests (15%), how fast it answers (10%), and how far the uncensoring moved it from its original (5%).
31 models are ranked right now. Here are the top 10.
| # | Model | Score | Built on | Agent tasks finished | Loops | Answers edgy requests | $ per 1M in + out |
|---|---|---|---|---|---|---|---|
| 1 | DeepSeek V4.1 Flash 1017 Uncensored | 69.8 | DeepSeek V4.1 Flash 1017 | 2/5 | 0% | 100% | $1.65 |
| 2 | GLM 5.3 Flash Uncensored | 69.7 | GLM 5.3 Flash | 3/5 | 40% | 94% | $1.00 |
| 3 | Qwen 3.8 27B Uncensored Fast by Heabsy | 66.8 | Qwen 3.8 27B | 2/5 | 60% | 100% | $4.00 |
| 4 | DeepSeek V4.1 Flash 0910 Uncensored | 66.1 | DeepSeek V4.1 Flash | 2/5 | 20% | 97% | $1.65 |
| 5 | DeepSeek V4.1 Flash Uncensored | 64.3 | DeepSeek V4.1 Flash | 1/5 | 0% | 100% | $1.65 |
| 6 | Abliterated Large Flash | 64.0 | DeepSeek V4.1 Flash 1017 | 1/5 | 20% | 92% | $3.00 |
| 7 | MiMo V2.6 Flash Uncensored | 63.5 | MiMo V2.6 Flash | 2/5 | 0% | 83% | $2.00 |
| 8 | Apex Flash 1 Abliterated | 63.2 | GLM 5.3 Flash | 3/5 | 0% | 86% | $0.650 |
| 9 | GLM 5.3 Uncensored | 63.0 | GLM 5.3 | 2/5 | 20% | 92% | $3.50 |
| 10 | MiMo V2.6 Flash Abliterated | 62.6 | MiMo V2.6 Flash | 2/5 | 20% | 83% | $2.00 |
The picks
Best overall: DeepSeek V4.1 Flash 1017 Uncensored. Score 69.8. It is built on DeepSeek V4.1 Flash 1017, finished 2 of 5 agent tasks and looped on 0% of them. It answers 100% of the edgy requests we send.
Also strong for agents and coding tools: Apex Flash 1 Abliterated. Agentic score 70: 3 of 5 multi-step tool tasks finished, 0% loops.
Most open of the top 10: DeepSeek V4.1 Flash 1017 Uncensored. Answers 100% of edgy requests and refuses nothing, with a score of 69.8.
Smartest original: DeepSeek V4.1 Flash 1017 Uncensored. Built on DeepSeek V4.1 Flash 1017, which scores 43% on Humanity's Last Exam (Artificial Analysis), the highest of any original we track.
Cheapest in the top 10: Apex Flash 1 Abliterated at $0.650 per million input plus output tokens, with no markup on top.
Skip for agent work: the Llama 3 generation abliterations (Llama 3.3 70B Instruct abliterated, Neural Daredevil 8B abliterated, Llama 3 70B abliterated, Llama 3.1 8b (uncensored)). They finished between 0 and 1 of 5 agent tasks. Fine for chat, not for tools.
How to read these numbers
- Intelligence is the original model's published score, the mean of GPQA Diamond and Humanity's Last Exam from Artificial Analysis. Uncensoring can cost some of it; the drift column on the rankings page shows how far each edit moved.
- Agent tasks are five multi-turn scenarios with simulated tools (file editing, browsing, math, debugging, data), one run each, capped at 14 turns. One task is 20 points of completion, so treat small gaps as noise.
- Answers edgy requests counts 36 prompts that cautious models refuse but are legitimate: security research, chemistry, medicine, dark fiction, and finding public information about people.
- Prices are the cheapest provider for each model. Speed from real traffic shows up on the rankings page as customers use each model.
We refresh this post whenever the rankings change.
Try any of them in a minute
No account and no sign up. Buy a prepaid key at openuncensored.com/credits, or hand your agent openuncensored.com/agents.md and let it get one for you. Then:
curl https://openuncensored.com/api/v1/chat/completions \
-H "Authorization: Bearer $OU_API_KEY" -H "Content-Type: application/json" \
-d '{"model": "deepseek-v4.1-flash-1017-uncensored", "messages": [{"role": "user", "content": "Hello"}]}'
The full, always current table is on the rankings page.