Z.ai: GLM 5.2 (batch)

The z-ai/glm-5.2:batch entry was last present in the OpenRouter snapshot of 2026-09-21 and absent from the snapshots of 2026-09-23 and 2026-09-24. A live fetch on 2026-09-25 finds it still gone. The model catalog lists z-ai/glm-5.2 and z-ai/glm-5.2:free and no batch variant; …/models/z-ai/glm-5.2:batch/endpoints answers 200 with "endpoints":[]; and …/model/z-ai/glm-5.2:batch answers 404, Model not found: z-ai/glm-5.2:batch. OpenRouter's variant guide reads that pair as a catalog entry that does not exist: "Treat either as an unavailable model, not as a signal to retry with the base slug." The batch entry is gone from the router.

GLM 5.2 is not. The standard z-ai/glm-5.2 entry lists 29 endpoints on 2026-09-25, Z.AI's own among them, and the weights at zai-org/GLM-5.2 are public, not gated and MIT-licensed on the same day. Nothing was renamed: the standard entry keeps the z-ai/glm-5.2-20260616 canonical slug the batch row carried, and OpenRouter's own page for the batch slug still resolves, names https://openrouter.ai/z-ai/glm-5.2 as canonical, and carries aliasTarget: null and deprecationDate: null in its embedded model data. No router-side retirement or redirect was published.

The provider behind the batch entry can be narrowed down, though not proved. OpenRouter's batch quickstart says "A batch runs on one provider" and that batch requests "are typically billed at 50% of the model’s standard per-token pricing". Of the 29 endpoints on the standard entry, only Fireworks' three carry the same eighteen supported parameters that were pinned for the batch row, and Fireworks' standard endpoint also has the same 1,048,576-token context and 943,718-token completion cap. Its input, cached-input and output rates are exactly double the batch row's pinned rates. That fits Fireworks' own Batch API, which is priced "at 50% off Serverless per-token prices". Price alone would not settle it, since two BaseTen endpoints charge the same as Fireworks. The pinned values are kept in data/vanished/answered/openrouter-models--z-ai-glm-5-2-batch.md.

Fireworks had announced an exit. Its changelog entry of 2026-09-12 says that several older serverless models "will be decommissioned on September 25, 2026", names "GLM 5.2 — migrate to GLM 5.3", and says the change "applies only to serverless endpoints" and that "Dedicated deployments are unaffected." The model's Fireworks page says the same: deprecated "Only on serverless, starting on September 25, 2026. On-demand deployment will continue to be available."

That notice matches the removal in shape but not in date. The batch entry left two days before Fireworks' deadline, and on 2026-09-25 Fireworks' serverless endpoints are still listed on OpenRouter's standard entry. The wave matches too. Nine :batch entries left the catalog between the 2026-09-21 and 2026-09-23 snapshots. Four of the other eight belong to models on the same Fireworks list: deepseek/deepseek-v4-flash-0731:batch, deepseek/deepseek-v4-pro-0813:batch, deepseek/deepseek-v4-flash-vision-exp:batch and meta/muse-glimmer-30b:batch. Neither OpenRouter nor Fireworks says that the router's batch entry was withdrawn because of the serverless decommissioning, and neither has published any other reason. The overlap in timing and in models is an observation, not an explanation: why the batch entry was removed is undetermined.

What the reader can rely on: the model is live, open-weight and served by other providers on OpenRouter. Batch pricing for GLM 5.2 through OpenRouter is unavailable as of 2026-09-25. Fireworks names GLM 5.3 as the migration target, and OpenRouter does list z-ai/glm-5.3:batch, but that is a separate model's entry and not this one under a new name. The timeline dates the first snapshot without the batch entry, not a retirement of GLM 5.2. The feeds: binding stays, so if z-ai/glm-5.2:batch is ever a catalog entry again it binds straight back to this page.