DeepSeek Launches V4-Flash API with Agentic Upgrades
DeepSeek has officially launched the API for DeepSeek-V4-Flash with major agentic capability improvements. According to the company’s benchmarks, the new version significantly outperforms V4-Pro-Preview in coding and autonomous agent tasks—scoring 54.4% on DeepSWE (near Opus 4.8) and 82.7% on TerminalBench. Key upgrades include enhanced long-horizon agent support, native Responses API, parallel tool calls, reasoning levels (low/high/max), a 1M token context, and compatibility with Codex via CLI and VS Code extensions. The model architecture and size remain unchanged from the preview version; only the API is updated for now, while V4-Pro API and web/app models stay unchanged.
Related: DeepSeek-V4-Pro-DSpark Released, DeepSeek Launches V4 with 1 Million Token Context