Models

Every request names the model that should answer it. This page covers how model IDs look, where to find a model's limits and capabilities, and what to expect from models that reason before they answer.

Model IDs

Model IDs follow the organization/model pattern, for example zai-org/GLM-5.3-Flash. IDs are case-sensitive, so copy them exactly as the API returns them.

There are three places to find the models you can use:

• GET /v1/models returns every model your API key can call, with its limits
• Models in the Entrim Dashboard lists the same models and shows code samples for each
• The model library on entrim.ai describes each model and lists its capabilities

A request with an unknown model ID fails with status 400 and a message that points you to /v1/models.

Context and output limits

GET /v1/models returns two limits for every model:

Field	Meaning
max_input_tokens	The largest input the model accepts: system prompt, conversation history, tool definitions and images together
max_output_tokens	The most tokens the model can generate in a single response

To cap a single response below the model's maximum, set max_tokens on the request.

Reasoning models

Some models work through a problem before they answer, and return that reasoning separately from the answer:

• The answer is in message.content, and the reasoning is in message.reasoning_content
• When streaming, the reasoning arrives as delta.reasoning_content chunks before the answer's delta.content chunks
• usage.completion_tokens_details.reasoning_tokens shows how many of the completion tokens went to reasoning

"message": {
  "role": "assistant",
  "content": "No, 221 is not a prime number. It factors as 13 × 17.",
  "reasoning_content": "221 = 13 × 17."
}

Reasoning tokens count toward max_tokens. If the limit is too low, the model can spend all of it on reasoning and return an empty content with finish_reason set to length, so leave plenty of headroom on reasoning models.

Use reasoning_effort to tell a reasoning model how much to think: low, medium or high. Lower effort answers faster and uses fewer tokens, while higher effort helps on harder problems.

You don't need to send reasoning_content back in later turns of a conversation. Add the assistant's content to messages as usual.

Tools, images and structured output

Models that support these features accept the same request fields as the OpenAI API:

• Tool calling: pass function definitions in tools and handle the tool_calls the model returns. See Tool calling.
• Images: send images as image_url content parts, from an https:// URL or as base64 data. See Images.
• Structured output: ask for JSON with response_format. See Structured output.

Check a model's page in the model library to see which of these it supports.