On July 31st, DeepSeek unveiled the public beta of the official API for DeepSeek-V4-Flash, emphasizing its significantly enhanced Agent capabilities. In a series of benchmark evaluations, V4-Flash delivered impressive results, securing 82.7 on Terminal Bench 2.1, 76.7 on the Cybergym cybersecurity assessment, 68.7 on the comprehensive full-stack development test DSBench-FullStack, 59.6 on the challenging coding test DSBench-Hard, and 25.2 on the ultimate Agent test. This performance closely rivals the 25.7 achieved by Opus-4.8 and far exceeds the 15.8 of the V4-pro preview edition. V4-Flash seamlessly integrates with the Responses API and aligns with the Codex ecosystem, making it straightforward for developers to incorporate into their projects. Regarding affordability, V4-Flash stands out with its high cost-efficiency, with 510,000 tokens costing a mere 0.53 yuan—a fraction of the cost compared to GPT-5.6 Sol, offering savings ranging from tens to hundreds of times. This update is specific to the V4-Flash API interface and does not impact the V4-Pro API, nor does it affect the APP and WEB versions. DeepSeek also disclosed that the official release of V4-Pro is on the horizon, promising further advancements.