[Daily Picks]Unsloth posted Dynamic v3.1 (preview) GGUFs for DeepSeek-V4.1-Flash-1017, arguing their quants keep more accuracy per size across several benchmarks (AIME25, LiveCodeBench, GPQA Diamond, and more). They also say they could not reproduce another lab’s GSQ-RCO recovery claims under the same settings, and publish averaged multi-run tables plus a Hugging Face GGUF pack. Lower-bit quants can still run with large unified memory (they mention ~128GB comfortably, or ~64GB with engram offloading).

Practical takeaway: when you pick a huge MoE/Flash GGUF, compare apples-to-apples—same eval suite, multiple runs, and watch reasoning length as well as score. “Smaller file” is not free if coding/agent benchmarks drop hard. Prefer publishers who show methodology and raw tables.

Which DeepSeek-V4.1-Flash quant are you actually running locally, and how does it feel on coding vs chat?

Source: https://www.reddit.com/r/LocalLLaMA/comments/1xjfuqh/introducing_deepseekv41flash1017_dynamic_v31/