DeepSeek V4 Pro delivers 5x better agent tasks with new APIs
DeepSeek's V4 Pro-0813 hits 62.7 on DeepSWE, crushing preview version's 12.8 score...
DeepSeek launched DeepSeek-V4-Pro-0813, the latest flagship V4 Pro model, now available on its website, mobile app and API. It scored 62.7 on the DeepSWE benchmark, up from 12.8 for the preview version, with gains in software engineering and agent-related tasks. The company also added Responses API and Codex integration for developers. Current API pricing is 3 yuan (~$0.42) per million input tokens and 6 yuan per million output tokens, with peak/off-peak adjustments starting August 17 — peak weekday output tokens will rise to 27 yuan per million, and off-peak will be half that amount.
- DeepSeek-V4-Pro-0813 scores 62.7 on DeepSWE (vs. 12.8 preview), enabling 5x better software engineering task resolution
- New Responses API and Codex integration expand agentic tooling for developers
- Pricing starts at $0.42/input token with peak-hour surcharges to $3.80/output token from Aug 17
Why It Matters
DeepSeek's V4 Pro accelerates enterprise automation by enabling AI agents to handle complex, multi-step tasks with minimal oversight, reducing operational costs.