Your privacy choices

Allow optional cookies for referral attribution, visit analytics, and Google Ads purchase measurement.

Back to blog
العربيةDeutschEnglishEspañolFrançais日本語한국어LatinaBahasa MelayuPortuguêsРусскийไทยTiếng Việt繁體中文中文

DeepSeek-V4-Pro-0813 Guide 2026: API, Codex & Price

DeepSeek-V4-Pro-0813DeepSeek V4 ProCodexResponses APIcoding agent

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

The exact keyword matters: DeepSeek-V4-Pro-0813 identifies the deployed model version, while API requests should normally keep using deepseek-v4-pro. On this gateway, that route is listed at 7.5x, approximately CNY 3 per million tokens, with a 1M context window.

Check the live route before a long agent task: review model pricing or start with a small balance on the buy page.

DeepSeek-V4-Pro-0813 key stats

Item Verified information
Current official version DeepSeek-V4-Pro-0813
Stable API model ID deepseek-v4-pro
Context window 1M tokens; Codex metadata uses 1,048,576
Maximum output 384K tokens
Default mode Thinking
Other mode Non-thinking
Official cache-hit input price $0.003625/M tokens
Official cache-miss input price $0.435/M tokens
Official output price $0.87/M tokens
Gateway multiplier 7.5x, about CNY 3/M tokens
Official interfaces Chat Completions, Responses, Anthropic
Image input Text-only in the current Codex metadata

Last checked: August 13, 2026. DeepSeek warns on its pricing page that overall API pricing is expected to increase in the near future, with the final plan subject to an official notice. Always recheck current prices before estimating production cost.

What does the 0813 version name mean?

DeepSeek-V4-Pro-0813 is the version label shown in DeepSeek's official Models & Pricing table on August 13. The public request name remains deepseek-v4-pro, allowing DeepSeek to update the version behind a stable API route without requiring every integration to change its configuration.

At the time of writing, DeepSeek's public change log does not include a separate August 13 launch note with architecture details or a before-and-after benchmark table for V4 Pro. The July 31 update explicitly covered DeepSeek-V4-Flash-0731 and said V4 Pro was unchanged at that time. The current pricing and Codex pages now confirm the 0813 version and Pro support, but they do not provide enough evidence to claim a specific percentage improvement.

That distinction is useful for SEO readers and production teams: the new version identifier is confirmed; unsupported performance claims are not.

Official DeepSeek V4 Pro features

The current official model table shows the same broad protocol coverage for V4 Pro and V4 Flash:

  • Thinking and non-thinking modes, with thinking enabled by default
  • JSON output and structured tool calls
  • OpenAI-compatible Chat Completions
  • OpenAI Responses API support
  • Anthropic-format API support
  • Chat prefix completion in beta
  • Fill-in-the-middle completion in non-thinking mode
  • 1M context and up to 384K generated output

DeepSeek's Codex guide describes V4 Pro as its most capable frontier agentic coding model. The model metadata marks it as text-only and exposes low, high, and max reasoning effort, with high as the default. Client and gateway support for every optional field can differ, so begin with the minimum payload and add controls one at a time.

DeepSeek-V4-Pro-0813 price explained

DeepSeek's direct API and this gateway use different billing systems:

Billing surface Cache-hit input Cache-miss input Output Notes
DeepSeek direct API $0.003625/M $0.435/M $0.87/M Current official list price
This gateway — 7.5x, about CNY 3/M Check live ledger Platform multiplier model

Do not convert one row into the other as if they were the same meter. Gateway account groups, output weighting, promotions, caching, and later price updates may change the billed amount. The live price page and actual account ledger are authoritative for this gateway.

DeepSeek also publishes a concurrency limit of 500 for V4 Pro in its current direct API table. That is an upstream direct-API limit, not a guaranteed per-account concurrency allocation on a third-party gateway.

Validate cost and compatibility together: make one short request after purchasing on the buy page, then confirm the route, token usage, response format, and ledger before increasing context or concurrency.

Call DeepSeek V4 Pro through the API

For normal OpenAI-compatible chat, use the stable model ID deepseek-v4-pro:

curl https://api.llm-token.cn/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 and list the rollback checks."}
    ]
  }'

For Codex and agent clients that use the Responses protocol, test /v1/responses separately:

curl https://api.llm-token.cn/v1/responses \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-pro",
    "input": "Inspect the repository and explain the failing test. Do not modify files."
  }'

Start with a short read-only request. A successful chat response does not prove that streaming, Responses events, tool calls, or a long-running Codex session will work identically. Verify each layer and retain the request ID when asking for support.

Configure DeepSeek V4 Pro in Codex

DeepSeek's official Codex instructions now include deepseek-v4-pro. A gateway configuration can use a custom Responses provider:

model_provider = "llm-token"
model = "deepseek-v4-pro"
model_reasoning_effort = "high"

[model_providers.llm-token]
name = "LLM Token"
base_url = "https://api.llm-token.cn/v1"
wire_api = "responses"
requires_openai_auth = true

Provide the API key through the secure authentication method supported by your Codex version. Do not save a real key in config.toml if that file is synchronized or committed. Begin with a read-only repository inventory, then separately validate search, patches, shell, tests, MCP, and long-context behavior.

The official model metadata reserves about 5% of the context for system and execution overhead. Treat “1M context” as the total model window, not as a guarantee that an entire 1M-token repository can be pasted as user input.

DeepSeek V4 Pro vs V4 Flash

Decision factor DeepSeek V4 Pro 0813 DeepSeek V4 Flash 0731
Positioning Most capable DeepSeek agentic coding route Lower-cost, faster agent route
Stable model ID deepseek-v4-pro deepseek-v4-flash
Context 1M 1M
Official output price $0.87/M $0.28/M
Gateway multiplier 7.5x 2.5x
Best starting point Difficult coding, planning, analysis High-volume or cost-sensitive agent work

Use V4 Flash as a cost baseline and move tasks to V4 Pro when the value of a more capable route is measurable. Compare both on the same repository, prompt, tools, timeout, and acceptance tests.

For a premium 500K-context alternative with a different coding and agent profile, read the Grok 4.6 API review.

Production checklist

  1. Send deepseek-v4-pro, not the version label, as the API model field.
  2. Confirm the returned version and route when the response exposes them.
  3. Test thinking defaults and reasoning controls with a short request.
  4. Validate Chat Completions and Responses independently.
  5. Set maximum output, timeout, tool-call, and total-cost limits.
  6. Keep destructive tools, deployment, payments, and infrastructure behind approval.
  7. Measure accepted-task rate, latency, token usage, and engineer rework.
  8. Recheck DeepSeek's official price warning before committing to a budget.

Frequently asked questions

What is DeepSeek-V4-Pro-0813?

It is the current official version name shown behind DeepSeek's V4 Pro API route. It is not the usual model string sent in API requests.

Which model ID should I send?

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

Does DeepSeek V4 Pro support Codex now?

Yes. DeepSeek's current Codex integration guide lists both V4 Flash and V4 Pro, and describes Pro as its most capable frontier agentic coding model.

How large is the DeepSeek V4 Pro context window?

The official model table lists 1M tokens. The Codex metadata uses 1,048,576 total tokens and a 95% effective-context percentage.

What is the DeepSeek-V4-Pro-0813 price?

DeepSeek currently lists $0.003625/M cache-hit input tokens, $0.435/M cache-miss input tokens, and $0.87/M output tokens. This gateway lists deepseek-v4-pro at 7.5x, approximately CNY 3/M tokens; verify the live ledger.

Is DeepSeek V4 Pro multimodal?

The current official Codex model metadata describes it as text-only, and this gateway marks image input unsupported. Use a dedicated vision model for image prompts.

Are there official 0813 benchmark improvements?

No separate public August 13 benchmark table was available in DeepSeek's change log when this article was checked. Avoid repeating percentage gains unless DeepSeek publishes a source.

Where can I buy or top up access?

New users can use the buy page. Existing users can use top-up.

Primary sources