DeepSeek v4 Flash final models show solid improvements

DeepSeek has launched the formal V4-Flash API in public beta, focusing on agent task upgrades. The model achieved 82.7 on Terminal Bench 2.1 and 54.4 on DeepSWE. This release integrates support for the Responses API and is adapted for Codex. DeepSeek reports that V4-Flash-0731 models show solid improvements. These updates aim to enhance performance across specific benchmarks, providing a more robust toolset for developers utilizing the latest DeepSeek flash model iterations.