Uncensored & abliterated Qwen
Yes. MagmaRouter sells an uncensored Qwen model: Qwen3.8 27B Xploded (qwen3.8-27b-xploded). Qwen3.8 Max and Qwen3.8 2.4T A95B and Qwen3.8 Flash Next and Qwen3.7 Plus are not sold uncensored.
Qwen, uncensored and abliterated, so far
Qwen is the most common base for community uncensored and abliterated builds, because the weights are strong at every size and published under a permissive licence. huihui-ai has released abliterated versions of most Qwen generations on Hugging Face, and "Qwen abliterated" usually means one of those. MagmaRouter sells Qwen3.8 27B uncensored: a finetune with the refusal training taken off rather than an abliteration, with half a million tokens of context and tool calling.
Qwen releases and the uncensored line
| Model | Intelligence | Uncensored on MagmaRouter | Context | Price per 1M (in / out) |
|---|---|---|---|---|
| Qwen3.8 Max | 45 | No | - | - |
| Qwen3.8 2.4T A95B | 40 | No | - | - |
| Qwen3.8 Flash Next | 40 | No | - | - |
| Qwen3.8 27B | 34 | Yes, as Qwen3.8 27B Xploded | 1M | $1.50 / $5.00 |
| Qwen3.7 Plus | 25 | No | - | - |
Intelligence is the Artificial Analysis Intelligence Index v4.3.2 for each base model, checked 2 October 2026. Uncensored builds are not benchmarked separately.
Calling Qwen3.8 27B Xploded
Any OpenAI client works with the base URL https://api.magmarouter.com/v1; see OpenAI-compatible API for the full set of examples.
curl https://api.magmarouter.com/v1/chat/completions \ -H "Authorization: Bearer sk-mr-..." \ -H "Content-Type: application/json" \ -d '{ "model": "qwen3.8-27b-xploded", "messages": [{"role": "user", "content": "Hello"}] }'
FAQ
Is there an uncensored Qwen API?
Yes: Qwen3.8 27B Xploded (qwen3.8-27b-xploded) on MagmaRouter, through an OpenAI- or Anthropic-compatible endpoint.
Is the Qwen on MagmaRouter abliterated or uncensored?
Qwen3.8 27B Xploded is uncensored: the refusal training was taken off, not edited out along the refusal direction.
Can it drive Claude Code or another coding agent?
Qwen3.8 27B Xploded calls tools, so yes. Every model on the line supports tool calling and answers both the OpenAI and Anthropic formats; setups are in the guides.