DeepSeek V4 Pro 0813: API Analysis & Benchmarks

The DeepSeek V4 Pro 0813 is a sophisticated, large-scale Mixture-of-Experts (MoE) architecture developed by DeepSeek. This model is currently available via OpenRouter for developers seeking high-capacity intelligence.

📊 Technical Specifications & Cost

The following table outlines the core economics and capabilities of the model:

Note on Actual Costs: While the listed prices are the baseline, the real-world expenditure is frequently lower. Due to standard listed rates caching mechanisms and various discounts, the effective price paid by the user is often significantly reduced.

---

🛠️ Infrastructure and Routing

Currently, this model is serviced by a single provider. Because there is only one source, OpenRouter does not need to perform complex routing logic; it simply forwards the request.

---

📈 Performance Metrics Explained

When evaluating this model, OpenRouter tracks several key indicators to determine its suitability for production environments:

• Throughput: Measured in tokens per second. $\text{Higher} \rightarrow \text{Better}$.
• TTFT (Time-to-First-Token): The latency between the request and the first character of output. $\text{Lower} \rightarrow \text{Better}$.
• Uptime: The success rate of requests over a rolling 30-day window.
  • Reliability: Higher percentages indicate a more stable model.
  • Rank Percentile: This shows how the model's stability compares to every other model on the platform.

OpenRouter ensures high availability by continuously monitoring providers and automatically attempting a retry with the next-best available option if an error occurs.

---

💻 Integration and Usage

The API is designed for seamless adoption, as it maintains full compatibility with the OpenAI standard. Most existing SDKs will function perfectly provided you update the base URL and the model identifier.

Model Slug: deepseek/deepseek-v4-pro-0813

Example Implementation:

import openai

client = openai.OpenAI(
    base_url="https://openrouter.ai/api/v1",
    api_key="YOUR_OPENROUTER_KEY",
)

response = client.chat.completions.create(
    model="deepseek/deepseek-v4-pro-0813",
    messages=[{"role": "user", "content": "Hello!"}]
)

❓ Feature Checklist

The following items are common points of inquiry for developers implementing this model:

• Does DeepSeek V4 Pro 0813 support tool calling?
• Does DeepSeek V4 Pro 0813 support structured outputs?