DeepSeek releases V4.1 Flash-1017, claiming far lower per-task cost
AI summary
DeepSeek released V4.1-Flash-1017, updating its website, app and API and publishing MIT-licensed weights, according to a Recode China AI digest. The multimodal model keeps a 1M-token context window, and DeepSeek reports improved coding and agent performance with shorter outputs and faster task completion, including 77.0 on DeepSWE v1.1 using Mini-SWE-V2. cnBeta, citing a Pareto curve published by DeepSeek itself, says per-task cost is 1/50 to 1/100 that of models such as Fable and Astra; these cost and efficiency figures are vendor claims and have not been independently verified.
In-depth insight
The release pairs MIT-licensed weights and a reported 1M-token context window with vendor-reported per-task costs far below those of rival models.