Latest model aliases
An alias such as openai/gpt-luna-latest points to a model in that family. LiteLLM maintains the target. As LiteLLM moves the alias to a newer model, your application keeps the same model string.
Use the full provider/model-latest name from the table below. You do not need to create an alias mapping.
Python SDK
Install litellm with pip install litellm and set OPENAI_API_KEY to your OpenAI API key.
from litellm import completion
response = completion(
model="openai/gpt-luna-latest",
messages=[{"role": "user", "content": "Say hello."}],
)
print(response.choices[0].message.content)
A successful call prints a greeting from the model behind the alias.
AI Gateway
For an existing gateway, set OPENAI_API_KEY in the gateway environment. Add this entry to your config.yaml under model_list, then restart the gateway with that configuration.
model_list:
- model_name: openai/gpt-luna-latest
litellm_params:
model: openai/gpt-luna-latest
api_key: os.environ/OPENAI_API_KEY
If your gateway already routes openai/* through a wildcard deployment, you can use this alias without adding a model entry.
Set LITELLM_PROXY_API_BASE to your gateway's base URL, such as http://localhost:4000, and LITELLM_PROXY_API_KEY to a LiteLLM key with access to this model. Send a request from your client machine.
curl "$LITELLM_PROXY_API_BASE/chat/completions" \
-H "Authorization: Bearer $LITELLM_PROXY_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"openai/gpt-luna-latest","messages":[{"role":"user","content":"Say hello."}]}'
A successful response contains a greeting in choices[0].message.content. The gateway's response model field can show the alias. LiteLLM calculates cost using the model that served the request.
Supported aliases
These aliases call each provider's own API.
| Provider | Model names |
|---|---|
| OpenAI | openai/gpt-luna-latest, openai/gpt-sol-latest, openai/gpt-terra-latest, openai/gpt-astra-latest, openai/gpt-mini-latest |
| Anthropic | anthropic/claude-opus-latest, anthropic/claude-sonnet-latest, anthropic/claude-haiku-latest, anthropic/claude-fable-latest |
| xAI | xai/grok-latest |
| Moonshot | moonshot/kimi-latest |
| Z.AI | zai/glm-latest, zai/glm-flash-latest |
| DeepSeek | deepseek/deepseek-pro-latest, deepseek/deepseek-flash-latest |
| DashScope | dashscope/qwen-max-latest, dashscope/qwen-flash-latest |
| MiniMax | minimax/minimax-latest |
Gemini, Mistral, and DashScope's qwen-plus-latest already have provider-managed latest names. LiteLLM passes those through. The aliases above do not extend to Azure, Bedrock, or other hosting providers.
See the alias table for targets. To control when you change models, use a specific model ID instead. A newer target can change pricing and model behavior.
