Chinese AI lab DeepSeek has made its V4-Pro model generally available. It is accessible via app, web and API, is built for AI agents and handles contexts of up to one million tokens.
Built for autonomous agents
The GA version (DeepSeek-V4-Pro-0813) focuses on agent tasks – using tools, executing code and completing multi-step workflows without human intervention. The model handles up to a 1 million token context and can produce up to 384,000 tokens of output; in either “thinking” or “non-thinking” mode.
Benchmarks
On agent-focused tests, DeepSeek reports V4-Pro scoring 87.9 on Terminal Bench 2.1 and 62.7 on DeepSWE, among others – benchmarks that measure solving real, multi-step tasks.
Higher prices
Alongside the launch, DeepSeek raised its API prices: output tokens rise at peak hours from 0.87 to 3.96 US dollars per million (effective 16 August). The model had been in preview since April.
Sources: DeepSeek API Docs, Quartz.
