LIVE MODELV4 FLASH / API MAP

The shortest path to a V4 Flash response

Call DeepSeek V4 Flash with model deepseek-v4-flash through MyDeepSeek's OpenAI-compatible /api/chat/completions endpoint; Claude Code can use /anthropic, while current Codex CLI uses the /v1 Responses base.

Context
1M
tokens
Max output
384K
tokens
Cache hit
$0.0028
per 1M input
Output
$0.28
per 1M tokens
Operational brief

From signal to shipping

01 / REQUEST

Make the first Chat Completions call

Send a bearer API key, the deepseek-v4-flash model ID, and a normal messages array. The gateway returns OpenAI-compatible content and usage fields for billing verification.

cURL · Chat Completions
curl https://mydeepseekapi.com/api/chat/completions \
  -H "Authorization: Bearer $MYDEEPSEEK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [{"role": "user", "content": "Return READY"}],
    "stream": false
  }'
02 / ROUTING

Match the base URL to the wire protocol

A base URL is not interchangeable across clients. Claude Code appends /v1/messages to the Anthropic base, Codex speaks Responses, and OpenAI-compatible SDKs append /chat/completions.

  • General SDK base: https://mydeepseekapi.com/api
  • Claude Code base: https://mydeepseekapi.com/anthropic
  • Codex provider base: https://mydeepseekapi.com/v1
03 / PRODUCTION

Verify usage, not only status 200

A production smoke test should confirm the served model, prompt/output token counts, cached-token detail when available, and the corresponding usage record. Stream consumers should also wait for the terminal usage event.

  • Use a dedicated key per tool or environment
  • Set bounded output limits for automated smoke tests
  • Treat cached input as a reported usage category, not an assumption
Questions / answers

The short answers that unblock a build.

What is the V4 Flash model ID?+

Use deepseek-v4-flash. The current published underlying version is DeepSeek-V4-Flash-0731.

Is the V4 Flash API OpenAI compatible?+

Yes for Chat Completions, and the model is also available through the Anthropic-compatible Messages and OpenAI Responses surfaces described here.

Does the API support streaming and tools?+

Yes. V4 Flash supports streaming and tool use; use the protocol-specific event and tool formats expected by your client.

Official DeepSeek sources

Ready / next action

Keep separate keys for local, CI, and agent traffic.

Get a production key