1-bit Compressed Version of DeepSeek-V4.1-Flash 763B Model Surfaces

A model dubbed DeepSeek-V4.1-Flash-1017 GSQ-RCO Coder, a 763B-parameter sparse MoE coding model, has surfaced online. ISTA DASLab's compressed version prunes half the experts and quantizes weights to 1 bit, reportedly retaining over 95% of performance.

2026-11-02 ~ 2026-11-03 · 2 related posts

• ISTA releases 763B DeepSeek coder at 1 bit per param, pruning half the experts with 95% performance kept — Loginhe · 2026-11-02
• Unverified 'DeepSeek-V4.1-Flash-1017 Coder' claims 763B params at 1-bit weights — solyarisoftware · 2026-11-03