본문으로 건너뛰기

DeepSeekLLM

DeepSeek V4 Pro API

Model IDdeepseek/deepseek-v4-pro

DeepSeek V4 Pro on the OpenAI-compatible chat completions API, with a 1,048,576 token context, function tools and JSON schema output.

  • 1.05M context
  • 도구
  • JSON

플레이그라운드에서 DeepSeek V4 Pro 실행

Send a chat request with your account and see the exact token charge. Code needs an API key.

매개변수

모델
model
deepseek/deepseek-v4-pro
messages[].content
messages[0].content
max_tokens
temperature

요청 미리보기

{"model": "deepseek/deepseek-v4-pro","messages": [],"stream": true,"max_tokens": 1024}

웹 Playground는 로그인된 세션으로 이 본문을 보냅니다. 코드에서는 같은 본문을 API 키와 함께 보내세요.

결과

Send a message to run DeepSeek V4 Pro with your account.

Usage and charge appear here after a reply.

DeepSeek V4 Pro 가격

USD per 1M tokens. The tier is chosen by the prompt size of each request.

가격 불러오는 중…
  1. ReserveWhen a request is accepted, Hermes reserves credits for the estimated prompt plus max_tokens of output (8,192 when you set none). Set max_tokens to keep the reservation small.
  2. SettleWhen the response finishes, the reservation is replaced by the exact charge from the usage the model reports: prompt tokens at the input rate, cached prompt tokens at the cache-read rate, completion tokens at the output rate. The remainder is released immediately.
  3. FailuresIf the model rejects the request nothing is charged. If a stream is interrupted after tokens were generated, the generated text is estimated and charged.

DeepSeek V4 Pro로 할 수 있는 일

DeepSeek’s V4 Pro release highlights agent workflows and adjustable reasoning effort. Manufacturer API features and settings may differ from the OpenAI-compatible Hermes route; consult its request documentation and the current model status before relying on a particular option.

DeepSeek V4 Pro is served through the OpenAI-compatible endpoint as deepseek/deepseek-v4-pro. The catalog lists a context window of 1,048,576 tokens, function tools and structured JSON output. Image input is not listed, so send text only.

Billing is per token from prepaid credits, with a separate, lower cache-read rate for cached prompt tokens.

활용 사례

  1. Structured extraction

    Ask for JSON that matches a schema with response_format.

  2. Tool-using agents

    Function tools pass through unchanged.

  3. Long text

    Use the listed context for long documents and logs, allowing space for the response and other request tokens.

코드에서 DeepSeek V4 Pro 호출

Any OpenAI SDK works: set the base URL and your Hermes key.

  1. 1

    API 키 생성

    In the dashboard, with the chat:create scope. API 키
  2. 2

    Point your SDK at Hermes

    Base URL https://api.hermes-ai.net/api/v1
  3. 3

    Send a chat completion

    POST /chat/completions with deepseek/deepseek-v4-pro

코드 예시

Set HERMES_API_KEY in your environment.

shell
export HERMES_API_KEY=your_api_key
Full chat API reference
curl https://api.hermes-ai.net/api/v1/chat/completions \
  -H "Authorization: Bearer $HERMES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v4-pro",
    "messages": [{"role": "user", "content": "Say hello in one sentence."}],
    "max_tokens": 200
  }'

Request

POST https://api.hermes-ai.net/api/v1/chat/completions API key with the chat:create scope, or your signed-in session.

필드유형참고
modelstringdeepseek/deepseek-v4-pro
messagesarrayOpenAI chat messages: system, user, assistant and tool roles; text and image_url parts.
streambooleanServer-sent events with a final usage chunk.
max_tokensintegerOutput cap; also sizes the credit reservation.
tools / tool_choicearray / objectFunction tools only.
response_formatobjectJSON mode and JSON schema where the model supports it.
temperature, top_p, stop, seed, reasoning_effortPassed through unchanged.

응답

필드참고
choices[].message / deltaThe OpenAI chat completion shape.
usageprompt_tokens, completion_tokens and prompt_tokens_details.cached_tokens as reported by the model.
hermes.charged_microsExact charge in USD micro-units (JSON responses). Streams expose it on GET /api/v1/chat/requests.
x-hermes-requestRequest id header, also used in the billing ledger.
error.codeDocumented Hermes error code; branch on it, not on the message.

DeepSeek V4 Pro API 자주 묻는 질문

Answers built from the model's catalog entry and the Hermes AI API contract.

Can DeepSeek V4 Pro read images?

The catalog does not list vision for this model. For image input, pick a model with vision from the LLM catalog.

Does it return JSON reliably?

The catalog lists structured output support, so you can send a JSON schema in response_format. Validate the result in your code all the same.

What is the DeepSeek V4 Pro API on Hermes AI?

DeepSeek V4 Pro by DeepSeek is served through the OpenAI-compatible chat completions endpoint, https://api.hermes-ai.net/api/v1/chat/completions, with the model ID deepseek/deepseek-v4-pro.

What is the context window of DeepSeek V4 Pro?

1,048,576 tokens, shared by the prompt and the output.

Does DeepSeek V4 Pro support vision, tools and JSON output?

Supported: 도구 및 JSON. Not supported: Vision, Reasoning 및 오디오. Streaming works with stream: true.

How is DeepSeek V4 Pro billed?

Per token, from prepaid credits. When a request is accepted, credits are reserved for the estimated prompt plus max_tokens of output (8,192 when you set none); when it finishes, the exact charge is taken from the usage the model reports and the rest is released. A rejected request is not charged. Rates are in the pricing table.

Can I use the OpenAI SDK with DeepSeek V4 Pro?

Yes. Set the base URL to https://api.hermes-ai.net/api/v1 and use your Hermes API key; send deepseek/deepseek-v4-pro as the model. Chat completions, streaming, tools and JSON output work unchanged.

What happens when I hit a rate limit?

Too many requests per minute return 429 RATE_LIMITED and too many in flight return 429 CONCURRENCY_LIMITED. If the model provider rate limits, you get 429 UPSTREAM_RATE_LIMITED and nothing is charged. Retry with backoff.

How do I get an API key for DeepSeek V4 Pro?

Create one under API 키 in the dashboard with the chat:create scope, then add credits on the Billing page.