Chinese AI company DeepSeek has officially released its flagship V4 Pro model with ending a nearly four-month preview period and bringing major upgrades to AI agent capabilities. The model, DeepSeek-V4-Pro-0813 is now live on the company's website, app and API.
Unlike standard AI models that simply answer queries of AI agents can break down complex tasks, use external tools, execute actions and adjust their approach based on results requiring less human oversight.
DeepSeek reported strong gains on agent-focused benchmarks. On DeepSWE which tests real-world software engineering skills, the model scored 62.7 up sharply from 12.8 for the earlier preview version. It also scored 87.9 on Terminal Bench 2.1 with solid results in cybersecurity-related tasks as well.
The release adds support for the OpenAI-style Responses API and Codex integration also giving developers more tools to build agent-based applications. The model also supports a 1-million-token context window and outputs of up to 384,000 tokens.
Alongside the launch, DeepSeek is raising its API prices in some cases by over 1000% while introducing peak and off-peak billing. The new rates take effect August 16. During weekday peak hours (9am–noon and 2pm–6pm Beijing time), output token prices will rise significantly, with off-peak rates set at half the peak price.
The launch reflects a broader shift in AI development, moving from simple question-answering tools toward systems capable of handling coding with multi-step tasks and autonomous decision-making. Industry analysts say AI agents are likely to play an increasingly central role in software development and other technical fields going forward.
