News summary36KrNov 2, 2026, 01:38 AM
DeepSeek Launches V4.1 Flash-1017 and Publishes an Efficiency Frontier
aiartificial_intelligence
DeepSeek formally launched the V4.1 Flash-1017 model. [[fact:fact-dkkk82hte]] It also published a Pareto frontier curve showing the model's efficiency and cost performance. [[fact:fact-xm4mv5w9p]] Token usage was more than 50% lower than in the previous version in most scenarios. [[fact:fact-yblrlozoy]] DeepSeek's DeepSWE v1.1 evaluation chart showed a highest Pass@1 rate of 77.0% for the new version. [[fact:fact-ora3unqmr]] The iteration left the external API calling method and public pricing system unchanged. [[fact:fact-o8ny3mzkz]] The article stated that the new version's comprehensive cost per task was one-hundredth to one-twentieth that of comparable models such as Fable and Astra. [[fact:fact-fv1hwno2p]]