# Unsloth Qwen3.8\-27B Quantizations – Dynamic v3.1 Update

- HackerByte page: <https://hackerbyte.net/en/stories/unsloth-qwen3-8-27b-quantizations-dynamic-v3-1-update-50491980>
- Original source: <https://unsloth.ai/docs/models/qwen3.8>
- Hacker News discussion: <https://news.ycombinator.com/item?id=50491980>
- Submitted to Hacker News: 2026-11-14T00:06:04.000Z

## AI-generated summary

Unsloth has released an update for its Qwen3.8\-27B model quantizations, introducing the Dynamic V3.1 algorithm. This update improves accuracy by 5\-10% across various benchmarks, including coding and agentic tasks, while maintaining similar file sizes. The update is available for most quants from 2\-bit to 6\-bit and can be run locally via Unsloth Desktop or llama.cpp.

## Key points

- Unsloth Dynamic V3.1 algorithm improves Qwen3.8\-27B quants by 5\-10% in accuracy.
- The update applies to most quantizations from 2\-bit to 6\-bit.
- New NVFP4 quants offer \~1.5x speed improvement over BF16 checkpoints with 92\-97% accuracy retention.
- Qwen3.8\-27B can run locally on setups requiring 17GB RAM/VRAM.
- Support for 256K context window and extended context up to 1M via YaRN.