deepseek-v4.1IMAGE
V4.1-Flash-1017 在 Terminal-Bench 超 GPT 5.6 Sol，便宜约 17 倍
称 V4.1-Flash-1017 在 Terminal-Bench 上超过 GPT 5.6 Sol 且便宜约 17 倍，比上一版 V4.1-Flash 提升 35%，靠的是在多个编码 agent harness 上做 RL 训练而非算法创新。
𝕏AID ADIL 👨💻
@HOTHEAD01TH12 天前收录原推发布于 2026.10.18
𝕏AID ADIL 👨💻 @HOTHEAD01TH原推发布于 2026.10.18
DeepSeek V4.1-Flash-1017 has released, beating GPT 5.6 Sol on Terminal-Bench while being ~17x cheaper. It improves the previous DeepSeek V4.1-Flash result by 35%, accomplished with "no algorithmic innovation" but instead by RL training it on multiple coding agent harnesses.
x.com/hothead01th/status/2111723966918271237采集于 2026.10.18 16:33