fix: add MiniMax and Z.ai subscription params

[guillaumegay13]

Summary

• Preserve provider-native uppercase MiniMax model IDs and allow uppercase characters in model IDs.
• Correct existing MiniMax M2.7 subscription row casing and add missing MiniMax subscription rows for M2, M2.1, M2.1-highspeed, M2.5, and M2.5-highspeed.
• Add missing Z.ai subscription rows for glm-4.5, glm-4.6, and glm-5, and align all Z.ai Coding Plan rows with the matching API-key parameter set (do_sample, response_format.type, sampling applicability).
• Update schema docs and tests for provider-native casing, plus a catalog regression that keeps Z.ai subscription/API-key param paths aligned for shared models.

Docs checked

• MiniMax Anthropic-compatible API: https://platform.minimax.io/docs/api-reference/text-anthropic-api
• MiniMax text API model/parameter reference: https://platform.minimax.io/docs/api-reference/text-post
• Z.ai Chat Completion API reference: https://docs.z.ai/api-reference/llm/chat-completion
• Z.ai GLM Coding Plan overview: https://docs.z.ai/devpack/overview
• Z.ai GLM Coding Plan quick start / Coding API endpoint: https://docs.z.ai/devpack/quick-start

Live smoke

• Used local provider keys from .env.keys; no secrets printed or committed.
• MiniMax model-list smoke passed: GET https://api.minimax.io/anthropic/v1/models?limit=100 returned all 7 Manifest-known subscription IDs.
• MiniMax completion smoke was attempted for all 7 rows with max_tokens, temperature, and top_p, but every call failed before model execution with upstream insufficient balance (1008).
• Z.ai Coding Plan model-list smoke passed: GET https://open.bigmodel.cn/api/coding/paas/v4/models returned all 7 Manifest-known subscription IDs.
• Z.ai completion smoke was attempted for all 7 rows with max_tokens, temperature, top_p, and thinking.type; extra probes for do_sample + response_format.type and thinking.clear_thinking were also attempted. All completion probes failed before model execution with upstream 429 / code 1309 because the Coding Plan package is expired.

Notes

• MiniMax official docs publish these language model IDs with MiniMax-* casing.
• MiniMax docs currently conflict on thinking: the support table says thinking is supported, but the later important note says thinking is ignored. Because live completion is blocked by account balance, this PR keeps MiniMax to the unambiguous scalar params already supported by the row shape.
• This same-repo branch supersedes fork PR fix: add MiniMax and Z.ai subscription params #30 so Vercel can run without fork authorization failure.
• Git may display two MiniMax YAML changes as high-similarity renames because the parameter rows are intentionally near-identical except for model IDs; the final tree contains the intended uppercase MiniMax files and no lowercase duplicates.

Validation

• npm run validate
• npm test
• npm run build
• npm run guard:params
• npm run typecheck
• npm run lint
• npx prettier --check on touched files
• git diff --check
• Direct Manifest MPS lookup smoke against the rebuilt local catalog: zero missing MiniMax/Z.ai subscription specs; MiniMax rows expose max_tokens, temperature, top_p; Z.ai rows expose max_tokens, do_sample, temperature, top_p, thinking.type, response_format.type.

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
modelparameters-dev	Ready	Preview	May 25, 2026 12:36pm