fix(ai-gateway): expand fallback model variants

[chrarnoldus]

Summary

• add fallback variants for popular Gemma, Laguna, Ling, and Nemotron model families
• align Claude max reasoning effort with the latest catalog metadata
• order OpenAI and custom LLM reasoning variants consistently from most to least intensive
• add focused coverage for family matching, endpoint-suffixed model IDs, and variant ordering

Validation

• pnpm --filter web test --runInBand src/lib/ai-gateway/providers/variants.test.ts src/lib/ai-gateway/custom-llm/order-opencode-variants.test.ts
• pnpm --filter web run typecheck
• pnpm exec oxlint --config .oxlintrc.json apps/web/src/lib/ai-gateway/providers/variants.ts apps/web/src/lib/ai-gateway/providers/variants.test.ts apps/web/src/lib/ai-gateway/custom-llm/order-opencode-variants.ts apps/web/src/lib/ai-gateway/custom-llm/order-opencode-variants.test.ts
• pnpm exec oxfmt --list-different apps/web/src/lib/ai-gateway/providers/variants.ts apps/web/src/lib/ai-gateway/providers/variants.test.ts apps/web/src/lib/ai-gateway/custom-llm/order-opencode-variants.ts apps/web/src/lib/ai-gateway/custom-llm/order-opencode-variants.test.ts
• git diff --check

All pnpm commands emitted the existing engine warning because the local runtime is Node 26 while the repository requires Node 24.

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Executive Summary

Incremental review of the Claude max-expectation test update verified the changed effort: 'max' assertions and updated catalog comment against REASONING_VARIANTS_CLAUDE (max → effort 'max' + verbosity 'max', xhigh → effort 'xhigh' + verbosity 'xhigh'), the isClaudeModel matcher, and the ReasoningEffortSchema/VerbositySchema enums with high confidence — the expectations now match the implementation, and no bugs, security issues, or memory-leak risks were found in the changed lines.

Files Reviewed (1 file)

• apps/web/src/lib/ai-gateway/auto-model/resolution.test.ts

Previous Review Summaries (2 snapshots, latest commit e0453e4)

Current summary above is authoritative. Previous snapshots are kept for context only.

Previous review (commit e0453e4)

Status: No Issues Found | Recommendation: Merge

Executive Summary

Incremental review of the variant-ordering alignment commit verified the reversed VARIANT_ORDERS rank lists and the OpenAI fallback .reverse() against the ReasoningEffortSchema option order, the isOpenAiModel matcher, updated tests, and all downstream consumers with high confidence — no bugs, security issues, or memory-leak risks found in the changed lines.

Files Reviewed (4 files)

• apps/web/src/lib/ai-gateway/custom-llm/order-opencode-variants.ts
• apps/web/src/lib/ai-gateway/custom-llm/order-opencode-variants.test.ts
• apps/web/src/lib/ai-gateway/providers/variants.ts
• apps/web/src/lib/ai-gateway/providers/variants.test.ts

Previous review (commit 515cab2)

Status: No Issues Found | Recommendation: Merge

Executive Summary

Fallback-variant additions for Gemma, Laguna, Ling, and Nemotron families plus the Claude max effort alignment were verified against ReasoningEffortSchema, provider matcher precedence, and the model catalog with high confidence — no bugs, security issues, or memory-leak risks found in the changed lines.

Files Reviewed (2 files)

• apps/web/src/lib/ai-gateway/providers/variants.ts
• apps/web/src/lib/ai-gateway/providers/variants.test.ts

---

_{Reviewed by kimi-k3 · Input: 67.9K · Output: 22K · Cached: 1.2M}

_{Review guidance: REVIEW.md from base branch main}