Models & API compatibility
Choose the right model ID and API format for your application.
On this page
Match your client to an APIChoose a model IDDeepSeek V4.1 FlashDeepSeek V4 ProCheck capabilities before enabling themCompare billing separatelyMatch your client to an API#
An endpoint supporting one format does not necessarily support the others. Confirm support before configuring a coding tool.
| Client | Required format | Typical request path |
|---|---|---|
| OpenAI SDK examples in this guide | Chat Completions | /chat/completions |
| OpenCode custom provider in this guide | Chat Completions | /chat/completions |
| Claude Code | Anthropic Messages | /messages relative to its Anthropic base |
| Codex | Responses | /responses |
Use the exact API base supplied by your provider. Do not append /chat/completions to an SDK's base URL: the SDK adds the request path.
Choose a model ID#
The official DeepSeek documentation currently lists deepseek-flash and deepseek-v4-pro. These are official API identifiers, not a promise that every independent service offers both models. DS-sub's model list has not been published.
A plan called Pro, Max, Pro Max or Ultra describes a subscription tier. It is not a model ID. In particular, a Pro plan does not imply access to the deepseek-v4-pro model.
DeepSeek V4.1 Flash#
The official API uses deepseek-flash for the current Flash model. The model family name shown in a client can differ from the identifier sent in a request. Check your provider's supported IDs before configuring it.
DeepSeek V4 Pro#
The official API uses deepseek-v4-pro. Availability, enabled capabilities and billing through an independent provider must be checked with that provider. A subscription tier named Pro does not select this model automatically.
Check capabilities before enabling them#
Confirm support for streaming, tools, structured output, images and reasoning settings separately. Start with a small text request; add optional parameters one at a time. Context limits and output limits depend on the model and provider.
Do not copy a context size or model alias from another provider into your coding assistant. Incorrect metadata can cause rejected requests or premature context compaction.
Compare billing separately#
Official API pricing is usage based. Subscription allowances and reward points follow the rules of their issuing service. Cached-input pricing on the official API does not automatically translate into the same discount on a subscription.
See pricing comparison, plan usage, and the official models and pricing reference. Verify current rates before estimating cost.
Guide updated November 5, 2026 · All tutorials