Quick answer
Mistral now hosts Z.ai’s GLM-5.2 as a third-party, public-preview text model under model ID zai-glm-5-2. Mistral says it serves the model without modifications and positions it for long-context coding and agentic workflows.
That creates a useful hosting option, not provider equivalence. Z.ai owns the model; Mistral owns the hosted endpoint, regional controls, service operation, billing surface, and platform terms. Use the regional readiness checker before selecting an endpoint.
Current provider record
| Field | Current documentation |
|---|---|
| Model provider | Z.ai |
| Hosting provider | Mistral |
| Mistral model ID | zai-glm-5-2 |
| Lifecycle | Public Preview |
| Modification statement | Hosted without Mistral modifications |
| Modality | Text |
| Context and output | 1 million context, up to 128,000 output tokens |
| Documented features | Chat completions, function calling, structured and predicted outputs, prefix, batching |
Preview status, availability, price, limits, and supported regions can change independently. Pin a source check to every production approval.
Model support is not endpoint support
The model card lists batching, but Mistral’s regional-inference documentation says regional endpoints do not support Batch, Agents, or the Files API. Function calling is currently the only supported regional tool.
Therefore a feature can be supported by the model or global API and still be unavailable on a selected regional endpoint. Query available models against the exact base URL and test every required feature there.
Read the regional feature limitations guide for alternatives and the EU/US data-location guide for control-plane boundaries.
Acceptance plan
Build a versioned test set for coding, long-context retrieval, structured output, function calls, refusal quality, latency, token use, and failure recovery. Record the model ID, endpoint hostname, request identifier, served service tier, region, timestamp, and current price source.
Do not assume that open weights, third-party hosting, or API compatibility establishes identical output, safety policy, retention, support, or uptime across Z.ai, Mistral, and self-hosted deployments.
Frequently asked questions
Is GLM-5.2 a Mistral model?
No. Mistral documents zai-glm-5-2 as a third-party open-source text model from Z.ai that Mistral hosts without modifications.
Is Mistral-hosted GLM-5.2 generally available?
The reviewed model card labels it Public Preview. Recheck lifecycle, model availability, region, price, limits, and terms before production.
Does the GLM-5.2 model card prove regional feature support?
No. Model capabilities and endpoint capabilities are separate. Mistral regional endpoints currently exclude Batch, Agents, and Files and support function calling as the only regional tool.
Official sources
- Mistral: Z.ai GLM-5.2 model card
- Mistral: Regional inference
- Mistral: In-region inference and open models
Source check: August 12, 2026. Recheck preview status, model revision, price, features, region, limits, and provider terms.