Unsloth Releases Dynamic V3.1 GGUFs for Qwen3.8-Flash-Next with 10-15% Accuracy Gains

·2026.11.05 03:08

Key point

New quantization files improve accuracy by 10-15% on agentic benchmarks and introduce a new engram switching process for multiple precisions.

Unsloth released Dynamic V3.1 GGUFs for Qwen3.8-Flash-Next, achieving 10-15% better accuracy on DeepSWE, Terminal Bench, and Aider Polyglot compared to GSQ-RCO.
The release includes a new engram switching process supporting BF16, Q8_0, Q5_1, and IQ4_NL, with file sizes ranging from 28.8 GB to 102.4 GB and RMSE values from 0 to 0.0757 relative to BF16.
Benchmarks were conducted using 256K context for agentic tasks and 128K for others, with a total of 26.2B tokens processed to mitigate noise, including 18.1B tokens for DeepSWE and 5.2B for Terminal Bench.
All quants smaller than UD-Q4_K_XL are currently updated, with larger quants planned for future updates; files are available on Hugging Face.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.