fix: match Z.AI parameter limits to provider docs

[brunobuddy]

💭 Why

Audit against https://docs.z.ai/guides/overview/concept-param showed our Z.AI entries were missing max_tokens bounds and glm-5.3 carried options the API rejects. Every change was verified against the live Z.AI API.

✨ What changed

• max_tokens gets default 65536 plus max 131072 (GLM-5/5.1/5-turbo/4.7/4.6) or 98304 (GLM-4.5 family) on 19 entries.
• glm-5.3 reasoning_effort trimmed to low/high/max (API rejects the rest with error 1210).
• glm-5.3 thinking can no longer be disabled ("This model always engages in thinking"), enum reduced to enabled.
• Regenerated npm package data.

📝 Notes

• Live API quirk: the gateway accepts max_tokens up to 131072 for glm-4.5 and glm-4.5-x, but the docs cap all GLM-4.5 output at 98304 (96K) and the air/airx/flash siblings validate at 98304. Kept the documented value.
• glm-4.7-flash/flashx are absent from the doc table; their 131072 limit comes from the API validation error.
• glm-4.7 still accepts thinking disabled (tested), only glm-5.3 forces it.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
modelparams.dev	Ready	Preview	Aug 22, 2026 6:13pm