Add sap z.ai GLM-5.2

[tjandy98]

No description provided.

[github-actions]

Action items

• [high] [violation] providers/sap-ai-core/models/zai--glm-5.2.toml:5 - Check: Reasoning options on this niche provider must document the exact request syntax and reflect SAP AI Core's actual API surface, not the upstream model's. Why: reasoning_options = [{ type = "effort", values = ["high", "max"] }] is copied verbatim from the zhipu/zai upstream entries, but SAP AI Core's provider.toml documents that orchestration v2 nests routed controls at $.config.modules.prompt_templating.model.params.<field> (e.g. reasoning_effort under params, not at request root). Every comparable sap-ai-core reasoning model (gpt-5, gpt-5.4, gpt-5.5, gemini-2.5-flash/pro, gemini-3.5-flash, claude-3.7-sonnet, claude-4.6/4.7-opus) carries a leading comment block stating the adapter request path and value mapping, and AGENTS.md requires this for niche providers. The new file has no such comment and no PR-body citation, so there is no evidence SAP accepts reasoning_effort=high|max for this model. Action: Add a leading comment block above base_model documenting the SAP AI Core request path for reasoning_effort and the high/max value mapping, with a first-party SAP/zhipu citation; confirm SAP actually forwards these values.
• [medium] [possible mistake] providers/sap-ai-core/models/zai--glm-5.2.toml:15 - Check: Provider pricing must be factual and supported. Why: cost.input = 0.00 and cost.output = 0.00 mark a flagship reasoning model as free on an enterprise orchestration platform where every other sap-ai-core model carries non-zero passthrough pricing (e.g. gpt-5.5 5/30, claude-4.8-opus 5/25, gemini-3.5-flash 1.5/9). The PR body is empty, so the $0 claim is unsupported. Action: Verify SAP AI Core pricing for GLM-5.2 against SAP documentation and set real per-million-token prices, or cite the SAP source that proves it is served at no charge.
• [low] [possible mistake] providers/sap-ai-core/models/zai--glm-5.2.toml:11 - Check: Provider limit overrides must be real provider limits, not placeholders. Why: The override sets context = 128_000 and output = 128_000 — identical round values that differ from the inherited upstream (context = 1_000_000, output = 131_072). An output limit equal to the entire context window leaves no room for input and looks like a copied placeholder rather than a measured SAP cap, and no citation supports either value. Action: Confirm SAP AI Core's actual context and output caps for GLM-5.2 against SAP documentation and cite the source; correct output if it should be lower than context.