Live now · beta round
Qwen 3.8 27B Uncensored
qwen38-27b-abliterated · served unquantized
The community's favorite uncensored model: tensor-level abliteration with the vision tower and thinking intact, zero measurable capability loss, and the speed that made it famous. Roleplay-tuned for the Roleplay tier, and sharp enough for agents, coding, and research on the upper window tiers.
Unquantized~100+ tok/s64K → 262K windowsVisionTool callingThinking
At a glance
- Parameters
- 27B dense, hybrid attention (GDN + full)
- Context window
- Ladder: 64K (Roleplay) → 128K (Plus) → 262K (Pro and above)
- Modality
- Text · vision · tool calling · thinking
- Decode speed
- ~100+ tok/s per stream
- Precision
- Unquantized, full training precision
- Uncensoring
- Tensor-level · 0% refusal on the A/B suite
- License
- Apache-2.0
- Endpoint
- api.unquant.io/v1 · OpenAI-compatible
Use it in one call
# OpenAI-compatible: point your SDK at us from openai import OpenAI client = OpenAI( base_url="https://api.unquant.io/v1", api_key="UQ-...", ) resp = client.chat.completions.create( model="qwen38-27b-abliterated", messages=[{"role": "user", "content": "Hello!"}], )
OpenAI-style image parts work on every tier: send an image alongside text and the vision tower does the rest.
Unlimited tiers on this model
| Tier | Window | Streams | Price | |
|---|---|---|---|---|
| Roleplay | 64K | 1 | $34.99/mo | Choose |
| Qwen Plus | 128K | 1 | $49/mo | Choose |
| Qwen Pro | 262K | 2 | $79/mo | Choose |
| Qwen Unlimited | 262K | 4 | $109/mo | Choose |
| Qwen Founding 46% off forever | 262K | 4 | $59/mo locked for life | Claim a seat |
| Qwen Dedicated | 262K | 8 · whole node | $549/mo | Talk to us |
All Qwen tiers include vision, tool calling, and thinking. Full tier list on the pricing page.