deepseek-v4.1POST
DeepSeek V4.1 Flash 在 DeepSWE 得 77.0，超过 Opus 5
帖子称 V4.1-Flash-1017 在 DeepSWE 上拿到 77.0，超过 Claude Opus 5，成本却低得多，并提到同周 Beam 501B 与 Step 5 Preview 开源。
ElizabethAI
@AISmartNest12 天前收录原推发布于 2026.10.18
ElizabethAI @AISmartNest原推发布于 2026.10.18
DeepSeek V4.1-Flash-1017 just scored 77.0 on DeepSWE — ahead of Claude Opus 5 — at a fraction of the cost. Same week: Reflection open-sourced Beam (501B) and StepFun dropped Step 5 Preview weights (600B). Open-weight agent models are closing the gap fast.
x.com/aismartnest/status/2111721048454615113采集于 2026.10.18 16:24