fix(alibaba): gate thinking_budget to qwen models, declare reasoningMaxTokens for kimi-k2.5

[AmineAce]

Summary

Declares reasoningMaxTokens: true for kimi-k2.5 on providerId: "alibaba" — a mapping left unverified in #3075 (cn-beijing region access issue).

Why a code change too

Just adding the catalog field wasn't safe on its own. DashScope's docs show enable_thinking works for Kimi, but thinking_budget is Qwen-only — sending it to Kimi is undocumented and untested. So thinking_budget is now gated to Qwen models only; enable_thinking stays unconditional.

Changes

• moonshot.ts: kimi-k2.5 (alibaba) → reasoningMaxTokens: true
• prepare-request-body.ts: thinking_budget only sent when externalId starts with qwen

Source

DashScope API reference (Chinese docs):

• enable_thinking: supports Kimi-K2.6/K2.5, Qwen, DeepSeek-V4, GLM
• thinking_budget: Qwen-only

Not changed

• No other Alibaba mappings affected
• glm-5 (also flagged unverified in fix: reasoning_effort for alibaba/minimax/xiaomi #3075) — separate scope
• Not live-tested (no cn-beijing API access) — docs-verified only

Testing

• pnpm exec turbo run build --filter=gateway — 10/10 pass
• pnpm format — clean

Summary by CodeRabbit

• Bug Fixes
  • Improved support for advanced reasoning with the Kimi K2.5 model through Alibaba.
  • High reasoning effort now correctly enables thinking mode and applies the appropriate token budget.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: a441612a-050f-4558-a5fd-13ba52e8f322

📥 Commits

Reviewing files that changed from the base of the PR and between a0bae80 and 82d48cb.

⛔ Files ignored due to path filters (1)

• pnpm-lock.yaml is excluded by !**/pnpm-lock.yaml

📒 Files selected for processing (1)

• packages/actions/src/prepare-request-body.spec.ts

---

Walkthrough

Alibaba’s Kimi K2.5 metadata now enables reasoning max-token handling, and the corresponding request-body test expects enable_thinking: true with a thinking_budget of 24576.

Changes

Alibaba reasoning configuration

Layer / File(s)	Summary
Reasoning budget mapping and model configuration packages/models/src/models/moonshot.ts, packages/actions/src/prepare-request-body.spec.ts	Alibaba’s Kimi K2.5 provider configuration adds reasoningMaxTokens: true, and the reasoning-effort test expects thinking enabled with a budget of 24576.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Possibly related PRs

• theopenco/llmgateway#3075: Adds related Alibaba reasoning-effort to DashScope field translation behavior.

Suggested reviewers: steebchen, ratchaw

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title matches the main change by calling out kimi-k2.5 reasoningMaxTokens, and the thinking_budget note is still relevant to the Alibaba behavior update.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}