DeepSeek has rolled out the formal release of V4 Pro to its API without changing the model name, adding significantly stronger agent capabilities along with support for Responses API and Codex integration. Pricing in the API documentation is listed at 3 yuan per million input tokens and 6 yuan per million output tokens, while cached input hits are priced at 0.025 yuan per million tokens. Benchmark data for the formal V4 Pro release showed a Terminal Bench score of 87.9, putting it close to Fable 5, which scored 88.0. The update points to a push toward more capable AI agent workflows, where models handle multi-step tasks and coding-related operations through external tools and structured interfaces.