Choosing a model
Which BlazeInfer model to use for chat, coding agents, vision and long context, and how model IDs and aliases work.
BlazeInfer serves a small, focused catalogue of self-hosted open-weight models. The right choice depends mainly on context window, tool calling and vision.
Model IDs and aliases
Chat model IDs look like Hugging Face repository names, lowercased:
vendor/model-name, for example zai-org/glm-5.3-flash. Some models also have
short aliases (for example glm-5.3-flash). Aliases are accepted in requests
and rewritten to the full model ID, so the response shows the full ID.
Use the full ID in production
GET /v1/models lists full model IDs only, and aliases may be moved to a newer
checkpoint over time. Use the full ID in anything you intend to keep running.
Context window and output limits
The console lists the context window for each model. Set max_tokens explicitly in your requests, and stream long
generations (see Streaming).