Models
The model Kourier.sh currently serves behind its compatible API endpoints.
kourier.sh currently exposes one model to customer API keys:
| Model | API model string | Maximum context length | Notes |
|---|---|---|---|
| DeepSeek V4 Flash | DSV4-Flash-0731 | 278,528 tokens (272K) | Live coding and reasoning model served by the current AutoRound deployment. |
Choosing the model
Use the exact identifier DSV4-Flash-0731 in OpenAI Chat Completions,
OpenAI Responses, Codex, and both Claude Code model variables. Bifrost does
not translate aliases such as deepseek-v4-flash; an unregistered model
identifier is rejected.
Context limits
DeepSeek V4 Flash has a maximum context length of 278,528 tokens (272K),
counting the prompt and generated output together. For requests that may
generate up to 16,384 tokens, keep the input at or below 262,144 tokens.
The prompt plus requested output must remain within the 278,528-token total.
Routing
The model currently routes through Kourier's internal
kourier-dsv4-autoround provider. Provider names are diagnostic metadata,
not client configuration: clients send the model string, base URL, and their
Kourier API key.
Access
New accounts start on Free with dashboard access. Redeeming an Omega activation code—or receiving an Omega assignment from an administrator—unlocks API-key creation with up to three concurrent requests.