AnalysisAI ModelsNovember 2, 2026

Unsloth and ISTA dispute DeepSeek-V4.1 GGUF quantization benchmarks

Read original source →reddit.com

Unsloth reports its DeepSeek UD 3.1 GGUFs beat GSQ-RCO by 10-20% on DeepSWE, Terminal Bench, Automation Bench, HLE and SciCode at 512K context, with 14 benchmarks run to 128K. ISTA's Loginhe says the comparisons don't match its DeepSeek-V4.1-Flash-1017 results and that it used standard settings for the 763B model.

2 sources

More stories today

Open the live feed