DeepSeek released the general-availability version of V4 Pro on August 13, 2026. The model had been available as a preview since April; the August update brought a new checkpoint to the API and to Expert Mode in DeepSeek's app and website.
Developers kept the same API model name, deepseek-v4-pro. The release also added native support for the OpenAI Responses API, including a setup path for Codex. V4 Pro and V4 Flash gained three reasoning-effort settings: low for simple requests, high for everyday agent work and max for harder tasks.
DeepSeek reports an 87.9 score on Terminal-Bench 2.1 and 62.7 on DeepSWE for the V4 Pro GA version. DeepSeek used its own agent test configuration for these measurements. A different coding harness or repository could produce different results.
Pricing changed alongside the launch. DeepSeek announced peak and off-peak API rates, with off-peak usage priced at half the peak rate from August 16, 2026, at 16:00 UTC.