DSV4.1-1017 + DSH is a notable upgrade over original 4.1. I've found it to be the 1st DeepSeek model I can use for all sort of research & SWE tasks. It's over 50% of OpenCode volume & a huge % of Open

What happened: A user reports that DSV4.1-1017 plus DSH is a notable upgrade over the original 4.1, saying it is the first DeepSeek model usable for research and SWE tasks, and that they paused ZCode, Kimi Code and OpenCode usage. Key numbers: ARC-AGI-2 78.1% at $0.16/task; ARC-AGI-1 94.5% at $0.06/task; ARC-AGI-2 scores rose from 61.4% (V4 Flash 0731) to 72.9% (V4.1 Flash) to 78.1% Why it matters: The user calls it the best among recently released Chinese models for SWE but says other areas still trail Kimi K3, and describes DeepSeek Flash as the cheapest baseline AI for routine tasks. Coverage: 4 outlets in 24h

A user reports that DSV4.1-1017 plus DSH is a notable upgrade over the original 4.1, saying it is the first DeepSeek model usable for research and SWE tasks, and that they paused ZCode, Kimi Code and OpenCode usage.

The user calls it the best among recently released Chinese models for SWE but says other areas still trail Kimi K3, and describes DeepSeek Flash as the cheapest baseline AI for routine tasks.