deepseek-v4.1IMAGE
2-bit V4.1-Flash on two RTX 3090s matches hosted quality on hard reasoning, thinks 2x longer
A 2-bit quant of V4.1-Flash-1017 on two used RTX 3090s matched the hosted full model's quality on a 26-problem hard reasoning eval, at roughly double the thinking time.
Aleksey Smolenchuk
@lxeArchived 12 days agoOriginally posted 2026.10.19
Aleksey Smolenchuk @lxeOriginally posted 2026.10.19
2 bit quant of DeepSeek-V4.1-Flash-1017 on two used RTX 3090s matches the quality of hosted full model on hard reasoning at the cost of thinking 2x longer Preliminary eval on 26 problems ran on @pocketmodelai Runner: https://github.com/lxe/dsx Model: https://huggingface.co/lxe/DeepSeek-V4.1-Flash-1017-dsx-2bit
x.com/lxe/status/2111978072803270784Archived 2026.10.19 12:01