Unsloth updates Qwen3.8-27B GGUFs with Dynamic v3.1 quants
Original title: Unsloth Qwen3.8-27B Quantizations – Dynamic v3.1 Update
Unsloth says the new quants gain 5–10% accuracy at similar size; 4-bit fits 16–19GB VRAM.
Unsloth released Qwen3.8-27B GGUFs using its Dynamic V3.1 preview algorithm, dated November 11, 2026. Qwen3.8-27B is a 27B dense model with vision and reasoning, a 256K context window, and Unsloth says it runs on about 17GB RAM or VRAM. Most 2-bit through 6-bit quants were updated; 3-bit UD3 was left unchanged. Unsloth reports 5–10% higher average accuracy versus its prior quants on agentic and general benchmarks and disputes competing GSQ-RCO comparisons.
Unconfirmed: Accuracy gains and the critique of ISTA and GSQ-RCO results are Unsloth’s own benchmark claims.
- 4-bit Qwen3.8-27B is listed for 16–19GB VRAM, including RTX 5080 and 4090.
- Context is 262,144 tokens and can extend to 1M via YaRN.
- Agentic benchmarks used 256K context; other benchmarks used 128K.
- Unsloth says 3-bit UD3 matched UD3.1, so those files were not replaced.
- AI models
- Qwen3.8-27B
From the original article
13 min read at unsloth.ai
Guide to running Qwen3. 8 quants including Qwen3. 8-27B on your local setup.
By Retro_Dev · unsloth.ai
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.