feat(ai-gateway): expose Terminal Bench model stats

[lambertjosh]

Summary

• Enrich the existing OpenRouter model catalog with optional Terminal Bench 2.0 completion and per-attempt cost metadata.
• Publish only active, non-stealth results with at least five attempts and a complete-attempt cost.
• Cache the compact benchmark map for five minutes and fail open to stale or empty data so stats lookup failures never block model selection.

Verification

• Not manually verified locally.

Visual Changes

N/A

Reviewer Notes

• Prefer deploying this catalog enrichment before the client PR, although either order is safe because the new field is optional.
• Metrics attach only to canonical OpenRouter catalog entries, with the existing safe outer kilo/ prefix fallback.

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Executive Summary

PR cleanly enriches the OpenRouter model catalog with optional Terminal Bench summary data, using a well-structured 5-minute in-process cache with fail-open semantics; no security, correctness, or performance issues found.

Files Reviewed (5 files)

• apps/web/src/lib/model-stats/terminal-bench.ts
• apps/web/src/lib/model-stats/terminal-bench.test.ts
• apps/web/src/lib/ai-gateway/providers/openrouter/index.ts
• apps/web/src/lib/organizations/organization-types.ts
• apps/web/src/tests/openrouter-models.test.ts

Notes (non-blocking)

• terminal-bench.ts:34-35: bench.avgAttemptCostUsd === null || bench.avgAttemptCostUsd === undefined is equivalent to bench.avgAttemptCostUsd == null — minor style nit, not worth changing.
• openrouter-models.test.ts:112-115: model could be undefined if 'some-other-model' is absent from mockOpenRouterModels, which would cause an unhelpful TypeError rather than a meaningful assertion failure. Consider adding expect(model).toBeDefined() before the terminalBench assertion for better failure messages.
• The terminalBenchFor mock in the integration test bypasses prefix-stripping logic, which is acceptable for the scope being tested (direct ID match), but the unit tests in terminal-bench.test.ts cover that path correctly.

Fix these issues in Kilo Cloud

---

_{Reviewed by claude-4.6-sonnet-20260217 · 553,311 tokens}

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