Mistral’s model catalogue now lists Z.ai GLM-5.3 as a hosted third-party model for long-context coding and agentic workflows. The page, dated 15 September 2026, says the service runs the model without Mistral modifications and exposes a 1M-token context window with 128K maximum output.[1]
Hosted by Mistral, made by Z.ai
Mistral’s page separates model identity from service provider. It labels GLM-5.3 as Z.ai’s third-party open-weight text model and says it is hosted by Mistral for long-context coding and agentic workflows. The same page says the model is served without Mistral modifications.[1]
The change sits at the access layer. Z.ai announced GLM-5.3 on 14 August and describes its gains over GLM-5.2 as post-training work. Its documentation lists text-only input, 1M context, 128K output and always-on reasoning with low, high and max effort.[2][3]
What Mistral exposes
The Mistral page is marked GA, Third-party, Open, v5.3. It lists chat completions, reasoning, function calling, structured outputs, predicted outputs and batching. Prices shown on the page are $1.40 per million input tokens, $0.14 per million cached input tokens and $4.40 per million output tokens.[1]
These fields describe Mistral’s serving offer. They do not establish a new checkpoint, a different training run or an independent quality result. The reviewed pages give no controlled test of the same GLM-5.3 through Mistral and Z.ai.[1][2]
Why the hosting route matters
Access through Mistral may simplify integration for teams already using its endpoints, particularly when a workload needs long context, reasoning, tool calls or structured output. The model remains text-only in Z.ai’s documentation, so the reviewed pages provide no basis for a multimodal reading.[1][2]
TechCrunch reported in October that Mistral had been hosting Chinese models while resisting the idea that it was becoming a mere inference provider. GLM-5.3 makes that product direction visible in the catalogue, while deployment conditions still need checking by account and region.[4][1]
What remains to verify
Before using the endpoint in production, readers still need current pricing, regional availability, retention and limits from Mistral’s live inference documentation. Those details can change independently of model weights. The current record supports one bounded claim: Mistral hosts Z.ai GLM-5.3 and says the served model carries no Mistral modifications.[1]
Sources
[1] Z.ai GLM 5.3 – Mistral AI | Mistral Docs
[2] GLM-5.3 – Overview – Z.AI DEVELOPER DOCUMENT
[3] GLM-5.3: Frontier Coding with Emergent Cyber Capabilities
[4] Mistral’s new 1T model aims to leapfrog closed and open rivals | TechCrunch
