'DeepSeek-V4-Flash-0731' API released... "Agent performance greatly improved through post-training"

DeepSeek has launched the public beta API for its V4-Flash-0731 AI model.

The model keeps the same architecture as V4 Pro Preview, but performance was boosted through re-applied post-training.

It showed major gains across agent benchmarks, along with better token efficiency and lower hallucination rates.

The API natively supports the OpenAI Responses API format, while pricing remains unchanged.

The update highlights how post-training alone can materially improve agent performance and intensify low-cost, high-performance AI competition.