Z.ai

Use the zai provider for Z.ai's native OpenAI-compatible Chat Completions API. The native catalog includes GLM-5.3, GLM-5.3-Flash, and GLM-5.2.

Configure

Set the environment variable:

export ZAI_API_KEY="your-api-key"

Or configure it in fast-agent.secrets.yaml:

zai:
  api_key: "${ZAI_API_KEY}"

Optional provider settings:

zai:
  api_key: "${ZAI_API_KEY}"
  base_url: "https://api.z.ai/api/paas/v4/"
  default_model: "glm-5.2"
  # default_headers:
  #   X-Custom-Header: value

The default endpoint is Z.ai's standard API. GLM Coding Plan users can select its Chat Completions endpoint without changing providers:

zai:
  api_key: "${ZAI_API_KEY}"
  base_url: "https://api.z.ai/api/coding/paas/v4/"

Models

Model	Native alias	Input	Context	Maximum output
GLM-5.3	`zaiglm53`	Text	1,000,000	131,072
GLM-5.3-Flash	`zaiglm53flash`	Text, image, linked MOV/PDF	1,000,000	131,072
GLM-5.2	`zaiglm`	Text	1,000,000	131,072

Use either an alias, bare canonical model ID, or provider-qualified ID:

fast-agent go --model zaiglm53 --message "Review this code."
fast-agent go --model glm-5.3 --message "Review this code."
fast-agent go --model zai.glm-5.3 --message "Review this code."

fast-agent go --model zaiglm53flash --message "Describe the attached image."
fast-agent go --model glm-5.3-flash --message "Describe the attached image."
fast-agent go --model zai.glm-5.3-flash --message "Describe the attached image."

For compatibility, the provider default and zaiglm alias remain GLM-5.2. The older glm and glm52 aliases also retain their Hugging Face routes:

zaiglm53       -> zai.glm-5.3
zaiglm53flash  -> zai.glm-5.3-flash
zaiglm         -> zai.glm-5.2
glm            -> hf.zai-org/GLM-5.2:zai-org
glm52          -> hf.zai-org/GLM-5.2:zai-org

GLM-5.3

GLM-5.3 is text-only. It supports streaming, function calling, context caching, separate reasoning_content, and JSON object structured output.

Reasoning is always enabled. Select one of the three documented efforts:

fast-agent go --model "zaiglm53?reasoning=low" --message "Solve this."
fast-agent go --model "zaiglm53?reasoning=high" --message "Solve this."
fast-agent go --model "zaiglm53?reasoning=max" --message "Solve this."

max is the default. GLM-5.3 rejects disabled thinking and unsupported effort values such as none and medium.

Sampling

Z.ai defaults GLM-5.3 sampling to temperature=1.0 and top_p=0.95. fast-agent leaves both fields unset unless requested, allowing the provider defaults to apply. Z.ai recommends tuning only one sampling control at a time:

fast-agent go --model "zaiglm53?temperature=0.8" --message "Write an introduction."
fast-agent go --model "zaiglm53?top_p=0.8" --message "Write stable technical documentation."

The same sampling defaults and recommendation apply to GLM-5.3-Flash.

GLM-5.3-Flash

GLM-5.3-Flash uses the same forced-reasoning contract and adds native multimodal input. fast-agent supports the model guide's JPEG and PNG inputs as remote URLs or Base64 data URLs, including multiple images:

fast-agent go --model zaiglm53flash

Then attach the images in the interactive prompt:

/attach https://example.test/first.png
/attach https://example.test/second.jpg
Compare these screenshots.

Z.ai limits images to JPG/JPEG/PNG, less than 5 MB each, and at most 6000×6000 pixels. WebP and GIF are not advertised for this profile.

The current Chat Completions API also accepts video_url and file_url content blocks for GLM-5.3-Flash. On August 26, 2026, fast-agent onboarding verified remote MOV (video/quicktime) and PDF links through the standard endpoint. These are therefore available as linked attachments. Embedded/Base64 video and document input, other video formats, and Office document input are not advertised without an equally specific contract.

Reasoning and tool continuations

Both GLM-5.3 models use preserved thinking:

{
  "thinking": {
    "type": "enabled",
    "clear_thinking": false
  },
  "reasoning_effort": "max"
}

fast-agent keeps hidden reasoning separate from visible assistant content. For later turns—including tool-result continuations—it returns the complete assistant reasoning_content unchanged and in its original order, as required by Z.ai.

Streaming channel normalization

Z.ai streaming deltas can carry reasoning_content, visible content, and tool_calls. fast-agent processes each field independently rather than treating them as mutually exclusive:

reasoning fragments remain in the structured reasoning channel;
visible content fragments remain in assistant content;
tool arguments are grouped and concatenated by tool-call index.

This preserves arrival order within each channel and supports alternating or same-chunk reasoning, content, and tool deltas. fast-agent does not merge hidden reasoning into visible assistant text or assume that the provider must finish all reasoning before emitting another delta type.

For streamed requests containing tools, fast-agent sends tool_stream: true. The non-streaming fallback removes only that streaming extension.

Structured output

All three native profiles use Z.ai's JSON object mode plus schema instructions and fast-agent validation:

fast-agent go --model zaiglm53 \
  --json-schema ./result.schema.json \
  --message "Return the requested result."

This is not strict OpenAI json_schema mode.

GLM-5.2 compatibility

GLM-5.2 retains its existing reasoning controls: none, minimal, low, medium, high, xhigh, and max. Unlike the 5.3 models, none disables thinking. GLM-5.2 remains text-only through the native profile.

Official documentation

See Models Reference for the generated capability row.