Back to Models

GLM 4.5V

zhipuzhipu/glm-4.5v

Zhipu AI GLM 4.5V β€” vision-language model supporting image, video, and document understanding. No function-call support.

GLM-4.5V is Zhipu AI's Z.ai visual reasoning model built on a Mixture-of-Experts architecture. Z.ai describes it as a 106B-parameter model with 12B active parameters, aimed at image, video, document, GUI, grounding, and chart-conversion tasks rather than plain text-only chat. On TheRouter it is published as zhipu/glm-4.5v, a multimodal chat-completions route that accepts text, image, video, and PDF inputs and returns text.

The practical distinction from newer GLM vision routes is important: standard-models.yaml marks this TheRouter route as having no function-call support. Use it when you need broad visual understanding and a 64K-context production route; move to zhipu/glm-4.6v-flashx when native tool calling or a lower per-token price matter more than staying on the 4.5V SKU.

Best for
  • β€’ Visual reasoning over mixed inputs: images, video, documents, screenshots, and PDF files in one OpenAI-compatible chat route
  • β€’ Document and chart interpretation, including summarization, translation, table extraction, and insight generation from multi-page reports
  • β€’ GUI and webpage screenshot analysis where the task is to describe layout, reconstruct interface code, or locate visible elements
  • β€’ Grounding and visual localization tasks that ask for target-object coordinates or bounding boxes in a scene
Reach for something else if
  • β€’ Agent pipelines that require function calling or tool_choice; TheRouter's zhipu/glm-4.5v route does not expose those controls
  • β€’ Pure text coding or math work; use a text-specialized GLM, Qwen, DeepSeek, or Kimi route instead of paying multimodal latency and price
  • β€’ Image generation or editing; GLM-4.5V is text-output vision-language, so route image creation to zhipu/cogview-4 or another image model

How TheRouter serves this differently from the vendor

As the vendor operates it

Z.ai documents GLM-4.5V as glm-4.5v on its own /api/paas/v4/chat/completions endpoint, including a thinking switch in examples and vendor-side multimodal usage patterns.

On TheRouter

TheRouter serves zhipu/glm-4.5v as an OpenAI-compatible chat-completions route with text, image, video, and PDF input, text output, 65,536 context tokens, 16,384 max completion tokens, and no function-call support in the published route contract.

Context Length
66K
Max Output
16K
Input Priceper 1M tokens
$0.648/ 1M tokens
Output Priceper 1M tokens
$1.94/ 1M tokens

Modalities

textimagevideopdf→text

Capabilities

Vision

Pricing Breakdown

TypeRate
Input$0.648 / 1M tokens
Output$1.94 / 1M tokens

Supported Parameters

temperaturemax_tokenstop_pstop

Specifications

ArchitectureMixture-of-Experts visual reasoning modeldocs.z.ai β†—verified
Parameter count106B total parameters; 12B active parametersdocs.z.ai β†—verified
Input modalitiesText, image, video, file / PDFdocs.z.ai β†—verified
Output modalityTextdocs.z.ai β†—verified
Maximum output tokens16K tokensdocs.z.ai β†—verified
TheRouter context window65,536 tokensdocs.z.ai β†—verified
Function calling on TheRouterNot supported on this zhipu/glm-4.5v routedocs.z.ai β†—verified
Thinking modeVendor API supports a thinking switch; TheRouter exposes the served route through its supported parameter listdocs.z.ai β†—verified
Training data cutoffNot publicly disclosedunknown
LicenseNot publicly disclosed for the hosted API routeunknown

Benchmarks

BenchmarkDistributionScoreSource
42 public visual multimodal benchmarks
Z.ai reports comprehensive performance across 42 public visual multimodal benchmarks but does not publish per-benchmark numbers in the retrieved guide.
β€”Not publicly disclosedβ€”
GUI tasks
The official guide names GUI tasks as part of the public benchmark coverage and gives GUI-agent examples, but no score table is exposed in the page text.
β€”Not publicly disclosedβ€”
Document understanding
Z.ai lists document understanding among the evaluated task families and separately describes complex long-document interpretation as a usage area.
β€”Not publicly disclosedβ€”

API Usage Examples

Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.

cURL
curl https://api.therouter.ai/v1/chat/completions   -H "Content-Type: application/json"   -H "Authorization: Bearer $THE_ROUTER_API_KEY"   -d '{
    "model": "zhipu/glm-4.5v",
    "messages": [
      {"role": "user", "content": "Summarize the key points from this input."}
    ]
  }'

Chat completion

Use this when the prompt is text-only or when you have already converted visual evidence into text.

cURL
curl https://api.therouter.ai/v1/chat/completions \
  -H "Authorization: Bearer $THEROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"zhipu/glm-4.5v","messages":[{"role":"user","content":"Summarize this document for a routing operator."}]}'

More from zhipu

Similar models

Cross-provider sibling models

News & changes

2026-08-15

Z.ai documents GLM-4.5V as a 106B / 12B-active visual reasoning model

The retrieved guide positions GLM-4.5V around multimodal reasoning, GUI work, grounding, chart conversion, and long-document interpretation. TheRouter keeps the route contract narrower than the vendor examples where unsupported parameters are not listed.

re-authored by TheRouterdocs.z.ai β†—

Frequently asked

Does zhipu/glm-4.5v support function calling on TheRouter?

No. TheRouter's published route contract for zhipu/glm-4.5v lists temperature, max_tokens, top_p, and stop, but not tools or tool_choice. Use zhipu/glm-4.6v-flashx if native multimodal tool use is required.

re-authored by TheRouterdocs.z.ai β†—
What inputs can GLM-4.5V handle?

The Z.ai guide lists video, image, text, and file inputs, and TheRouter publishes zhipu/glm-4.5v with text, image, video, and PDF input modalities. Output is text only.

re-authored by TheRouterdocs.z.ai β†—
When should I choose GLM-4.5V instead of GLM-4.6V-Flash?

Choose GLM-4.5V when you specifically need the older 4.5V route and its broad multimodal coverage. For new agentic vision workloads, GLM-4.6V-Flash is usually the sharper default because it is free on TheRouter and exposes native function calling.

re-authored by TheRouterdocs.z.ai β†—
Fact ledger β€” every claim on this page traces here
sourceURLretrieved
Architecturedocs.z.ai β†—2026-08-15verified
Parameter countdocs.z.ai β†—2026-08-15verified
Input modalitiesdocs.z.ai β†—2026-08-15verified
Output modalitydocs.z.ai β†—2026-08-15verified
Maximum output tokensdocs.z.ai β†—2026-08-15verified
TheRouter context windowdocs.z.ai β†—2026-08-15verified
Function calling on TheRouterdocs.z.ai β†—2026-08-15verified
Thinking modedocs.z.ai β†—2026-08-15verified
Training data cutoffβ€”β€”unknown
Licenseβ€”β€”unknown
42 public visual multimodal benchmarksdocs.z.ai β†—2026-08-15unknown
GUI tasksdocs.z.ai β†—2026-08-15unknown
Document understandingdocs.z.ai β†—2026-08-15unknown
Z.ai documents GLM-4.5V as a 106B / 12B-active visual reasoning modeldocs.z.ai β†—2026-08-15verified
Does zhipu/glm-4.5v support function calling on TheRouter?docs.z.ai β†—2026-08-15to verify
What inputs can GLM-4.5V handle?docs.z.ai β†—2026-08-15to verify
When should I choose GLM-4.5V instead of GLM-4.6V-Flash?docs.z.ai β†—2026-08-15to verify
Help & contact