Community releases 1-bit DeepSeek-V4.1-Flash coder

A community release packages a Coder build of DeepSeek-V4.1-Flash-1017, using GSQ quantization and RCO optimization to prune each layer from 384 to 192 experts and shrink the 551.9B transformer parameters to 72.1GB of weights, about 1.05 bits per parameter. It scores 88.57 on LiveCodeBench v6.

• Hidden persona mode found in DeepSeek-V4.1-Flash2026-10-18
• Free Chinese AI Models Approach OpenAI's Top Cyber Model in Hacking2026-11-06
• Report: closed models dominate price-performance frontier as open share hits low2026-11-06
• Delta Inference Engine v0.93 boosts local DeepSeek-V4.1 runs2026-11-05
• DeepSeek receives CNY 100 million from Tencent and CATL2026-11-05