Deepseek-V4.1-Flash-1017 tops DeepSWE at $0.20 per task

Deepseek's update to its Flash line is notable less for raw benchmark gains than for what it says about the cost curve. The company describes the release as a Pareto improvement—better scores, faster task completion, and shorter outputs all at once—and the DeepSWE cost chart backs that up, with the new model undercutting rival models that land between $2 and over $10 per task in the same chart.

The technical bet behind the model is reinforcement learning with self-feedback (RLSF), a post-training method that uses the model's own confidence scores or self-generated critiques as a reward signal, removing the need for human evaluators or external reward models. Deepseek says this helps end-to-end project development, from game programming to 3D modeling.

Separately, Deepseek is opening its in-house search engine—which it says already handles hundreds of millions of queries per day—to outside developers. That move comes as the Chinese government reportedly plans to tighten rules on publishing open-weight models, with one deciding factor being whether a system ranks among the top 10 most capable models. How that rulemaking unfolds could affect how freely Deepseek and others continue to release weights.