Models

fremai serves a curated catalog of frontier open models behind one OpenAI-compatible endpoint. You reference a model by its id in the model field — the same drop-in field the OpenAI API uses. GET /models returns the ids available to your key.

Because models are referenced by name, moving a model between backends is invisible to your code — the id names the model, not the host.

Launch line-up

Model	id (example)	From	Notes
GLM 5.2	`glm-5.2`	Zhipu AI	Flagship general + coding model. Strong tool-use and long-context reasoning.
MiniMax M3	see `GET /models`	MiniMax	High-throughput model built for fast, cost-efficient agentic workloads.
DeepSeek V4	see `GET /models`	DeepSeek	Frontier reasoning + coding model; a strong default for hard technical tasks.
Kimi 2.7	see `GET /models`	Moonshot AI	Long-context specialist for large documents and codebases.

The exact id strings, the embedding-model ids, and any additions beyond the launch line-up are returned live by GET /models and listed on the www.fremai.eu models page.

Context windows and pricing

Context windows and per-token prices are published per model at launch — see the fremai price page for the authoritative, per-model figures. Pricing is transparent and per-token; you pay on the backend-reported token usage returned in every response.

The catalog will grow beyond the launch line-up. New models appear in GET /models and on the price page as they are provisioned.

Choosing a model

General + coding, tool use: GLM 5.2 is a strong default.
Hard reasoning / technical tasks: DeepSeek V4.
Large documents / long context: Kimi 2.7.
High-throughput / agentic loops: MiniMax M3.

You can hold multiple keys scoped to different model subsets — e.g. a narrow CI key and a full interactive key.

Listing models programmatically

curl https://api.fremai.eu/v1/models \
  -H "Authorization: Bearer $FREMAI_API_KEY"

{
  "object": "list",
  "data": [
    { "id": "glm-5.2", "object": "model", "owned_by": "fremai" }
  ]
}