deepseek-v4.1POST
DeepSeek V4.1-Flash-1017 open weights: Terminal-Bench 31 to 42
V4.1-Flash-1017 ships MIT-licensed weights with the same ~550B/8B-16B-active architecture, Terminal-Bench up 31 to 42 and DeepSWE 77 above Opus 5; the checkpoint is about 510GB.
𝙒𝙞𝙡𝙘𝙤 𝘷𝘢𝘯 𝘽𝙤𝙭𝙩𝙚𝙡 ⚡🚀🧲🌞
@nuonrgArchived 10 days agoOriginally posted 2026.10.17
𝙒𝙞𝙡𝙘𝙤 𝘷𝘢𝘯 𝘽𝙤𝙭𝙩𝙚𝙡 ⚡🚀🧲🌞 @nuonrgOriginally posted 2026.10.17
DeepSeek just dropped V4.1-Flash-1017 with open weights (MIT). It has the same architecture as V4.1 Flash, roughly 550B backbone with 8B/16B active, but posts big agentic gains: Terminal-Bench 31 to 42 and DeepSWE 77, ahead of Opus 5. The 1017 checkpoint is about 510GB, while the existing V4.1 Flash Spark EXL3 recipe already runs on 2 DGX Sparks, so a local version should follow fast. Re-quant time. #AI https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017
x.com/nuonrg/status/2111378578998735230Archived 2026.10.18 16:45