DeepSeek in Claude Code: Official Model Names and Relay Setup Checks
For DeepSeek’s official API, use deepseek-flash as the Flash model request name. DeepSeek documents that it is now served by DeepSeek-V4.1-Flash-1017. For LLMAPI, select the exact request ID from the applicable model list or channel instead: do not treat an official vendor alias as proof of relay availability or Claude Code compatibility.
Keep the endpoint, model ID, protocol and key type together when configuring a connection. DeepSeek’s official API documentation and LLMAPI’s catalog describe different services; neither should be used as a substitute for the other’s configuration instructions.
Understand the official DeepSeek name mapping
The DeepSeek API documentation distinguishes the name you send in a request from the model serving it:
| Name | Meaning on DeepSeek’s official API |
|---|---|
deepseek-flash |
The instructed Flash request name, currently served by DeepSeek-V4.1-Flash-1017 |
deepseek-v4-flash |
A legacy request name that is still accepted; requests are served by DeepSeek-V4.1-Flash-1017 and billed at the Flash price |
deepseek-v4-flash-vision-exp |
Another accepted legacy name with the same documented replacement and Flash billing |
| DeepSeek-V4.1-Flash-1017 | The serving-model designation; use deepseek-flash as instructed rather than substituting this designation into the request |
The models corresponding to the two legacy names have been retired. Acceptance of an old name therefore does not mean that the old model is still serving the request. DeepSeek says the calling method remains unchanged.
For a new official Flash configuration, follow the documented deepseek-flash name. When reviewing an older configuration, check whether it uses a legacy alias before drawing conclusions about which model it calls. Refer to the official documentation for current pricing information rather than attaching an old price to an old alias.
Separate official direct access from LLMAPI access
DeepSeek documents OpenAI- and Anthropic-compatible API formats and says Claude Code can use DeepSeek directly as a backend. That is an official direct-access path, not a statement about which models a third-party relay exposes.
The documented base URLs are:
| Connection | Protocol | Base URL |
|---|---|---|
| DeepSeek official API | Anthropic | https://api.deepseek.com/anthropic |
| DeepSeek official API | OpenAI | https://api.deepseek.com |
| LLMAPI Claude Code connection | Anthropic | https://llmapi.pro — no /v1 |
| LLMAPI OpenAI-compatible connection | OpenAI | https://llmapi.pro/v1 |
Use the official DeepSeek documentation for direct access. Use the LLMAPI Claude Code guide for this site’s connection. Do not combine the official DeepSeek endpoint with an LLMAPI key, or carry a vendor model alias into a relay configuration without checking the relay’s request ID.
LLMAPI explicitly treats model names, protocol compatibility and billing mode as separate settings. For Claude Code, check Anthropic-protocol support for the selected model and channel; an OpenAI-compatible listing alone is not the check to use.
Configure the LLMAPI connection, then verify the model separately
The LLMAPI Claude Code guide documents these environment variables for macOS and Linux:
export ANTHROPIC_BASE_URL=https://llmapi.pro
export ANTHROPIC_AUTH_TOKEN=YOUR_API_KEY
For Windows PowerShell:
$env:ANTHROPIC_BASE_URL="https://llmapi.pro"
$env:ANTHROPIC_AUTH_TOKEN="YOUR_API_KEY"
Replace YOUR_API_KEY with the appropriate LLMAPI key from Console → API Keys. These snippets configure the service connection; they do not select a DeepSeek or Qwen model or establish access to one.
Before choosing a model:
- Open the model selection guide and identify the required billing mode.
- For subscription access, check the subscription model list. LLMAPI documents that this list is synchronized with
GET /v1/models. - For pay-as-you-go access, check the channel catalog for the request ID, pricing and supported protocols.
- Copy the request ID, not a display name, and confirm that the selected route supports the client protocol.
- Use the matching key type before launching the client.
Apply the same selection checks if you are considering Qwen. Do not choose a Qwen request ID or assume Claude Code support from the family name alone. This is a configuration checklist, not a tested DeepSeek or Qwen relay configuration.
Match billing and access before a longer session
LLMAPI documents two separate access modes:
- Monthly subscription: use a subscription key; supported models share the plan quota.
- Pay-as-you-go API: use an API billing key and a separate balance, with billing at the selected channel’s prices.
Topping up API credit does not increase subscription quota. Likewise, do not assume a model is covered by a subscription merely because it appears in a catalog or a client supports a compatible protocol. Check the exact subscription request ID or the applicable pay-as-you-go channel.
Use the current subscription plans and API channel catalog when comparing costs. DeepSeek’s official statement that legacy aliases are billed at the Flash price applies to its official API; do not use it as an LLMAPI channel price.
As a suggested final check, start with a short, non-sensitive text task before attempting a longer coding session. Record the intended service, exact request ID, protocol and billing mode without recording the secret key. Review the result and account usage before proceeding. Treat that check as validation of your own configuration, not as a benchmark or a guarantee of broader model capabilities.