deepseek-v4.1IMAGE
Hands-on: V4.1 Flash agentic coding jumps to domestic flagship level
The tester says 1017's agentic coding backend jumped from roughly GROK-4.6 level to domestic flagship level, but the model turned more conservative, needs a solid harness, and controls budget worse unless reminded each session.
karminski-牙医
@karminski3Archived 11 days agoOriginally posted 2026.10.20
karminski-牙医 @karminski3Originally posted 2026.10.20
同步下 deepseek-v4.1-flash-1017 的测试结果, 后端 Agentic Coding 能力直接从GROK-4.6差不多的水平跃升到国产旗舰模型水平了. (v4.1-pro 真的还有必要吗?) 不过我测下来 Agent 能力变化很大, 细节不是几百字就能描述清楚的, 简单来讲的话, 上一个版本很"主动", 而这个版本更"保守", 必须做好harness才行, 而且控制预算的能力好象变差了, 但如果每次会话强调预算, 它又会变好...... 详情预计本周会发个视频跟大家聊, 目前还在写POV, 证明猜想的一些问题.
x.com/karminski3/status/2112362936899616823Archived 2026.10.20 12:02