Qwen3.8 27B Uncensored GGUF Released with 262k Context and Vision Support
BLUECOW009 · x · 2026-08-23
The Qwen3.8-27B-Unleashed model has been updated with uncensored weights, vision capabilities, and MTP (Multi-Token Prediction) support.
Key Updates:
- Added vision component mmproj-Unleashed-f16.gguf.
- Retained MTP heads (nextn. tensors) for enhanced reasoning.
Performance & Specs:
- VRAM: 13GB (Q3 quantization).
- Speed: 100 tok/s on 1x 4090.
- Context: 262k (needle verified at 250k depth).
- Benchmark: MMLU 82.98% (measured on the 13GB quant).
The release utilizes Unsloth's Dynamic v3 recipe, which allocates bits based on tensor sensitivity (e.g., 8-bit for critical attention gates, 2-bit for less sensitive layers), allowing this 3-bit file to outperform standard 4-bit quants on perplexity.
Related event: Uncensored Versions of Qwen3.8-27B Emerge on Hugging Face(3 posts)→
More from Models
- Nathan Lambert slams Meta's Muse Open License as 'trash' — natolambert · 2026-10-16
- Google launches Gemini 4 Argon to Ultra and Pro subscribers, touting physics and 3D skills — ShikharMurty · 2026-10-16
- Google gives X users early access to test Argon, not just companies — lyraxana · 2026-10-16
- Grok Voice 3.0 Tops Design Arena Speech Leaderboard, Up 7.4 Points — XFreeze · 2026-10-16
- Robotics LLMs tested on 16 tasks: GPT-6 Astra hits 17.2% success vs Fable's 9.1% — pabbeel · 2026-10-16
- Meta open-sources Muse Spark: top US open model, scores just below DeepSeek — bindureddy · 2026-10-16