DeepSeek V4 Pro Reaches GA With Stronger Agents
The 0813 build upgrades agent performance, adds Responses API support and extends reasoning controls across DeepSeek services.
From preview to production release
DeepSeek has moved V4 Pro into general availability with the new 0813 build, rolling it out across its web application, mobile apps and API. Existing API users can access the update through the unchanged deepseek-v4-pro model name, so applications may receive the new build without altering their integrations.
The release concentrates on agent workloads rather than changing the model family’s underlying scale. DeepSeek’s published figures continue to describe V4 Pro as a mixture-of-experts model with 1.6 trillion total parameters, 49 billion activated per token and a one-million-token context window. The 0813 update reports substantially stronger results on software-engineering and tool-use evaluations, including 87.9 on Terminal-Bench 2.1, 74.1 on Toolathlon-Verified and 62.7 on DeepSWE. As always, vendor-reported benchmark results require independent reproduction, particularly for long-running agents whose reliability depends heavily on the surrounding software harness.
A more direct route into coding agents
DeepSeek’s API now supports the OpenAI Responses API format natively and provides configuration intended for Codex-compatible environments. V4 Pro and V4 Flash also gain three reasoning-effort settings—low, high and max—giving developers a way to trade latency and token consumption against deeper computation without switching models.
The company said revised peak and off-peak API pricing will take effect on August 16, with off-peak usage priced at half the peak rate. No newly post-trained 0813 weights were publicly available when the rollout began; this is therefore an official hosted-model update, not a fresh open-weight release.
The launch matters because DeepSeek is competing for the execution layer around coding and research agents, not merely benchmark rank. Native Responses API compatibility, long context and selectable reasoning effort reduce the work required to substitute DeepSeek for another provider. If the reported agent gains hold up in independent testing, V4 Pro could exert renewed price pressure on frontier agent models while giving developers a more credible non-US option for sustained tool-based work.