DeepSeek V4 Pro is the production-grade flagship from DeepSeek, the lab behind the open-weight V-series. The 0813 build left preview on August 12, 2026 and is now the stable model behind the deepseek-v4-pro API endpoint, aimed at developers and enterprises running long-context coding and agentic workloads.
Core Features
- Mixture-of-experts architecture with 1.6 trillion total parameters and 49 billion active per token.
- One-million-token context window with up to 384,000 tokens of output.
- Three operating modes: non-thinking, high reasoning effort, and max effort (V4-Pro-Max).
- Native tool calling, JSON output, and a Responses API, with compatibility for both OpenAI ChatCompletions and Anthropic Messages formats.
Use Cases / Best For
- Agent builders needing cheap, long-context reasoning with native tool use.
- Enterprises wanting open weights (MIT license) they can self-host.
- Developers already on OpenAI or Anthropic SDKs who want a drop-in alternative.
Pricing
API pricing is $0.435 per million input tokens on a cache miss, $0.003625 per million on a cache hit, and $0.87 per million output tokens. Open weights are free to download under the MIT license.
Our Take: Best for cost-sensitive agent workloads that still need frontier-class coding — it trails GPT-5.4 and Gemini-3.1 on some benchmarks but at a fraction of the price. Trade-off: vendor-reported benchmarks are not yet independently replicated.
Browse more in our AI Models & Engines category.
