DeepSeek API quickstart
Model IDs and access
Use an exact ID returned by your key. The public catalog does not guarantee live capacity or every optional capability.
deepseek/deepseek-v4-flashdeepseek/deepseek-v4.1-flashdeepseek/deepseek-v4-pro
Unprefixed compatibility IDs remain accepted; prefer the namespaced IDs returned by discovery.
export QUOTAFLOW_API_KEY="YOUR_QUOTAFLOW_API_KEY"
curl https://api.quotaflow.ai/openai/v1/models \
-H "Authorization: Bearer $QUOTAFLOW_API_KEY"
First request
OpenAI SDK base URL: https://api.quotaflow.ai/openai/v1. Keep your key on your server.
curl https://api.quotaflow.ai/openai/v1/chat/completions \
-H "Authorization: Bearer $QUOTAFLOW_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek/deepseek-v4.1-flash",
"messages": [
{
"role": "user",
"content": "Design an order-management API and explain its main endpoints."
}
],
"max_tokens": 4096,
"stream": false
}'
Read choices[0].message.content. For a plain-text conversation, send earlier user and assistant messages in messages; tool conversations also need the tool calls and matching results.
Responses clients
curl https://api.quotaflow.ai/openai/v1/responses \
-H "Authorization: Bearer $QUOTAFLOW_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek/deepseek-v4.1-flash",
"input": "Design an order-management API and explain its main endpoints.",
"max_output_tokens": 4096,
"store": false,
"stream": false
}'
Responses is stateless: send the full conversation in input every time. Do not use previous_response_id, conversation, prompt, background, or store: true. Keep returned reasoning and tool-call items intact when replaying them, and include matching tool results.
Compatibility and output budgets
Use the exact V4/V4.1 ID returned by discovery. Do not substitute historical deepseek-chat or deepseek-reasoner names for these models. Reasoning output, tools, streaming and JSON mode depend on the model/request; the family name alone proves none of them.
Use the output cap for the endpoint you call; these fields are not interchangeable. If an answer ends at its output limit, inspect its status/stop reason before increasing the budget within the model limit. For streaming, consume the endpoint-specific events until completion; do not assume a first chunk is the final answer.
401 indicates an authentication problem; check key access on 403. Correct invalid request fields on 400 instead of retrying unchanged. Use bounded backoff for temporary 429/503 capacity failures. Check model access with the same key and base URL; no guide guarantees live supply or a universal price.
See Responses API, Chat Completions and billing.