DeepSeek Wants Better AI at Lower Cost

What happened: DeepSeek released V4.1-Flash-1017, an update to its V4.1 Flash model, and open-sourced DeepInfer, its internal inference framework, plus deepseek-recipe, a library for processing API requests. Key numbers: Terminal-Bench 4.0 rose from 31.2 to 42.3; DeepSWE v1.1 rose from 74.2 to 77.0. Why it matters: DeepSeek is targeting efficiency at both the model and infrastructure levels, with shorter outputs and faster task completion aimed at lowering inference costs. The company notes its benchmark figures are its own and do not establish equal performance across all applications. Coverage: 4 outlets in 24h

DeepSeek released V4.1-Flash-1017, an update to its V4.1 Flash model, and open-sourced DeepInfer, its internal inference framework, plus deepseek-recipe, a library for processing API requests.

DeepSeek is targeting efficiency at both the model and infrastructure levels, with shorter outputs and faster task completion aimed at lowering inference costs. The company notes its benchmark figures are its own and do not establish equal performance across all applications.