DeepSeek V3 API: Fireworks Deployment and First Request

Short answer: This guide covers the Fireworks-hosted DeepSeek V3 route, not a DeepSeek-direct API call. Fireworks lists the model path accounts/fireworks/models/deepseek-v3; for a dedicated deployment, the request uses the deployment ID returned during creation. The listing is for the original DeepSeek V3—not V3.1 or V3-0324. Check that this route is available to your account before proceeding. (Fireworks DeepSeek V3 listing)

Prepare the Fireworks deployment

The Fireworks listing provides the base model path accounts/fireworks/models/deepseek-v3. Fireworks’ deployment workflow uses a separate deployment ID for inference; do not substitute the base model path in the request’s model field when calling a dedicated deployment. (Fireworks deployment guide)

1. Create a Fireworks API key. Install and sign in to firectl by following the official CLI instructions and on-demand quickstart. The default deployment shape requires firectl 1.8.8 or newer.

2. Check cost and lifecycle settings before creating capacity. Fireworks on-demand deployments use dedicated GPUs and are billed by GPU-second for active instances. Charges may accrue while an instance is active even when it is not handling API requests. Confirm the selected shape’s current rate and how you will manage its lifecycle before creating it. The V3 listing does not provide a numeric deployment price. (Fireworks deployment billing guide; billing and scaling FAQ)

3. Follow the Fireworks on-demand quickstart to create the deployment using the V3 model path above and a deployment shape. The quickstart describes default as having Fireworks select a validated shape for the model. Use the exact creation syntax and options in the current CLI documentation; save the deployment ID it returns, which follows the form accounts/<ACCOUNT_ID>/deployments/<DEPLOYMENT_ID>. The account’s available shapes and deployment eligibility may vary. (Fireworks quickstart; deployment guide)

Send a chat-completions request

Fireworks documents this endpoint for chat completions: https://api.fireworks.ai/inference/v1/chat/completions. Authenticate with a Fireworks API key as a Bearer token, send JSON, and set model to the deployment ID returned above. (Fireworks text-model guide)

export FIREWORKS_API_KEY="your_api_key"
export FIREWORKS_DEPLOYMENT_ID="accounts/YOUR_ACCOUNT_ID/deployments/YOUR_DEPLOYMENT_ID"

curl -X POST https://api.fireworks.ai/inference/v1/chat/completions \
  -H "Authorization: Bearer ${FIREWORKS_API_KEY}" \
  -H "Content-Type: application/json" \
  -d "{\
    \"model\": \"${FIREWORKS_DEPLOYMENT_ID}\",\
    \"messages\": [\
      {\"role\": \"user\", \"content\": \"Reply with a short greeting.\"}\
    ]\
  }"

In a successful chat-completions response, the generated text is in choices[0].message.content. Check the HTTP status and response body; this example describes the documented request pattern, not an authenticated test. (Fireworks text-model guide)

Production checks for this route

• Confirm the model and deployment: Use the Fireworks V3 listing and the deployment ID returned by the create workflow. Do not treat that provider-specific path or ID as a DeepSeek-direct model identifier.
• Estimate cost before deployment: On-demand capacity is billed by GPU-second for active instances, and an active instance may incur charges without API calls. The V3 listing does not give a numeric rate; check current pricing for the selected shape and its lifecycle settings before creating it.
• Check workload fit: The Fireworks listing reports a 131k-token context length and marks function calling as supported. These are claims for this Fireworks listing; confirm current details and the function-calling contract against provider documentation before relying on them. (Fireworks DeepSeek V3 listing)
• Protect the API key and monitor requests: Keep the key in an environment variable or secrets manager. Monitor errors and usage, and consult Fireworks’ deployment guide for current capacity and lifecycle guidance.

DeepSeek-direct access is a separate route

Do not send Fireworks model paths or deployment IDs to DeepSeek’s API. DeepSeek’s current first-call guide names deepseek-flash and deepseek-v4-pro; the available documentation here does not establish a direct DeepSeek V3 request contract. That documentation gap is not confirmation that direct V3 access is unavailable. Check DeepSeek’s current model list for available identifiers. (DeepSeek first-call guide)

Bottom line: For the Fireworks-hosted original V3 listing, create a dedicated deployment using Fireworks’ documented workflow, then call its chat-completions endpoint with the returned deployment ID. Confirm the current rate and lifecycle settings before creating capacity; this is not a DeepSeek-direct V3 integration.

Sources

DeepSeek V3 API & Playground | Fireworks AI

firectl CLI Getting Started | Fireworks AI Docs

Deployments Quickstart | Fireworks AI Docs

Deployments | Fireworks AI Docs

How does billing and scaling work for on-demand GPU deployments? | Fireworks AI Docs

Text Models | Fireworks AI Docs

Your First API Call | DeepSeek API Docs

Lists Models | DeepSeek API Docs