DeepSeek-V4-Pro-0813 API Guide: Codex, Responses & Pricing

What Is DeepSeek-V4-Pro-0813?

DeepSeek-V4-Pro-0813 is the current official version behind the stable API model ID deepseek-v4-pro. DeepSeek's latest documentation confirms a 1M token context window, up to 384K max output, thinking and non-thinking modes, JSON output, tool calling, Responses API, Anthropic API, and official Codex integration.

The two identifiers serve different purposes: DeepSeek-V4-Pro-0813 is the deployment version label, while deepseek-v4-pro is the stable route used in all API requests. Check the model pricing page for current rates and verify with a small request before running long tasks.

Verify routing and pricing before long tasks: Check model rates on the pricing page or open a small balance to confirm the actual route, tokens, and billing.

DeepSeek-V4-Pro-0813 Key Specs

What Does the 0813 Mean?

DeepSeek-V4-Pro-0813 is the version name shown on DeepSeek's Models & Pricing page as of August 13. The public request name remains deepseek-v4-pro, letting DeepSeek update the model behind the stable route without requiring client configuration changes.

As of this writing, DeepSeek's public changelog has no standalone August 13 V4 Pro architecture notes or benchmark comparison table. The July 31 update explicitly targeted DeepSeek-V4-Flash-0731 and stated V4 Pro was unchanged at that time. While the latest pricing and Codex pages confirm the 0813 version and Pro support, there is no authoritative source claiming a specific performance improvement percentage.

Official V4 Pro Capabilities

DeepSeek's current model table lists these capabilities for V4 Pro:

• Thinking and non-thinking modes, thinking enabled by default
• JSON output and structured tool calling
• OpenAI-compatible Chat Completions
• OpenAI Responses API
• Anthropic-format API
• Beta Chat Prefix Completion
• FIM mid-completion in non-thinking mode
• 1M context, up to 384K output tokens

DeepSeek's Codex integration docs describe V4 Pro as its most capable frontier agent programming model. Model metadata offers low, high, and max reasoning effort tiers, defaulting to high. Not all clients and gateways support every optional field, so start with minimal requests and add parameters incrementally.

Pricing Overview

DeepSeek's official direct API and this platform use different billing systems. Official DeepSeek API pricing is listed for reference only. Our platform uses a multiplier-based model — see the model pricing page for current rates. Actual costs depend on account grouping, output weights, promotions, caching, and future adjustments. Always confirm with the live pricing page and your account billing.

DeepSeek's official API also lists a 500 RPM concurrency limit for V4 Pro. This is an upstream limit on the official direct API and does not guarantee the same per-account concurrency on third-party platforms.

Calling DeepSeek V4 Pro via API

Use the stable model ID deepseek-v4-pro for standard OpenAI-compatible chat:

curl https://jzstoken.com/api/v1/chat/completions \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "deepseek-v4-pro", "messages": [ {"role": "user", "content": "Design a safe migration plan with rollback checks."} ] }'

For Codex and other agent clients using the Responses protocol, test /v1/responses separately:

curl https://jzstoken.com/api/v1/responses \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "deepseek-v4-pro", "input": "Inspect the repo and explain test failures without modifying files." }'

Start with short, read-only requests. A successful Chat Completions call does not guarantee identical behavior for streaming, Responses events, tool calling, or long-running Codex tasks. Validate each layer separately and keep request IDs for support.

Configuring DeepSeek V4 Pro in Codex

DeepSeek's latest Codex docs list both deepseek-v4-pro and V4 Flash. Through this gateway, configure a custom Responses provider:

model_provider = "jzstoken" model = "deepseek-v4-pro" model_reasoning_effort = "high" [model_providers.jzstoken] name = "JZS Token" base_url = "https://jzstoken.com/api/v1" wire_api = "responses" requires_openai_auth = true

Provide your API key through a secure authentication method supported by your Codex version. If config.toml is synced or committed, never store real keys in it. Start with read-only repo inventory tasks, then separately validate search, patch, shell, test, MCP, and long-context operations.

Official model metadata reserves roughly 5% of context for system and execution overhead. "1M context" refers to the total model window, not the full amount available as user input.

V4 Pro vs V4 Flash

Establish a cost baseline with V4 Flash first. Upgrade to V4 Pro only when the capability gain justifies the cost. Keep repo, prompt, tools, timeout, and acceptance tests fixed when comparing.

Production Integration Checklist

1. Send deepseek-v4-pro in the model field, not the version name.
2. If responses expose version and route, verify they match expectations.
3. Use a short request to confirm default thinking mode and reasoning effort parameters.
4. Test Chat Completions and Responses separately.
5. Set max output, timeout, tool calling, and total cost limits.
6. Destructive tool, deployment, payment, and infrastructure operations require approval.
7. Track task acceptance rate, latency, tokens, and engineer rework time.
8. Reconfirm DeepSeek's official pricing before budgeting.

FAQ

What is DeepSeek-V4-Pro-0813?

It is the current version label for the DeepSeek V4 Pro API route. It is typically not the model string used in API requests.

Which model ID should I use in the API?

Use deepseek-v4-pro. The service maps this stable route to DeepSeek-V4-Pro-0813.

Does DeepSeek V4 Pro support Codex?

Yes. DeepSeek's latest Codex integration guide lists both V4 Flash and V4 Pro, positioning Pro as the most capable agent programming model.

What is the context window?

The official model table lists 1M tokens. Codex metadata uses 1,048,576 as the total window with a 95% effective context ratio.

How much does it cost?

Check the model pricing page for current rates on this platform. Official DeepSeek API pricing is listed for reference only and may differ.

Does it support images?

Current official Codex metadata describes it as text-only. Use a dedicated vision model for image tasks.

Has DeepSeek published 0813 performance improvements?

No standalone August 13 benchmark comparison has been published. Do not share unsubstantiated performance claims without an official source.