DeepSeek-V4-Flash Official API Public Beta Launches with Million-Token Context and MoE Architecture

DeepSeek launched the official V4-Flash API public beta on July 31, 2026. The model, named deepseek-v4-flash, natively supports the Responses API format and is specifically adapted for Codex.

Facts

According to the official changelog, the DeepSeek-V4-Flash official API is now in public beta. The model structure and size of DeepSeek-V4-Flash-0731 remain identical to DeepSeek-V4-Flash-preview, with only post-training redone. The official DeepSeek-V4-Pro version will be released as soon as possible. V4-Flash supports open-source weights, MIT license, million-token context, and MoE architecture.

Mechanism Breakdown

This update applies only to the V4-Flash API; the V4-Pro API and APP/WEB models remain unchanged. The API invocation method is unchanged—simply set the model name to deepseek-v4-flash to use the latest version. The official blog disclosed that V4-Flash scored 82.7, 54.2, etc., on benchmarks such as Terminal Bench 2.1.

Industry Impact

For developers, integrating V4-Flash into tools such as OpenCode Go significantly improves code generation and agent task performance, while the MoE architecture with million-token context reduces long-text processing costs. For enterprise users, the MIT-licensed open-source weights facilitate private deployment, and the Responses API format's adaptation to Codex lowers the migration barrier.

Strategic Assessment

Based on available facts, the official V4-Pro release may further enhance enterprise-grade application capabilities. Developers can prioritize testing V4-Flash's stability in Codex-adapted scenarios to verify its performance in real-world workflows.