DeepSeek-V4-Pro
Language modelPrevious generationOpen weightsCard summarizedThe official release of DeepSeek-V4-Pro, with greatly enhanced agentic capabilities
Active (superseded by V4.1 architecture; phase-out paused)● Re-checked 12 Oct 2026 · 62 facts · 2 corrections · 1 unconfirmedSource: DeepSeek-V4-Pro-0813 model card
Use it via API
Where to call it
- DeepSeek API
deepseek-v4-pro - Self-host (weights)
deepseek-ai/DeepSeek-V4-Pro-0813
Good for
- Agentic coding and terminal tasks (Terminal Bench 2.1 87.9%)
- Knowledge-heavy tasks: strongest DeepSeek base model on SimpleQA-Verified and FACTS Parametric
- Long-context work up to 1M tokens
Overview
V4 Pro
The official release of DeepSeek-V4-Pro, with greatly enhanced agentic capabilities
- Cost
- $1.32 in
- Context
- 1M
- Input / output
- $1.32 / $3.96
- Max output
- 384K
- Knows up to
- —
- Released
- 13 Aug 2026
The GA release of DeepSeek-V4-Pro, superseding the April preview. Same 1.6T-parameter MoE structure (49B active, 1M context) with a DSpark speculative decoding module added and much stronger agentic results in production use. model card (0813)
deepseek-v4-pro- Input / output per 1M
- $1.32 / $3.96
- Context window
- 1M
- Max output
- 384K
- Knows up to
- —
- Reasoning
- Configurable (thinking on by default; low / high / max)
- Weights
- Open · MIT
About this card. A model card is the maker's own account of a model: what it's for, how it was trained and tested, and where it falls short. This page condenses DeepSeek-AI's DeepSeek-V4-Pro-0813 model card. Every claim carries its source page.