Update CrossRouter Submission

[JiaHg]

Update CrossRouter Submission

[JiaHg]

/evaluate

[github-actions]

Router Evaluation Results

Router: cross-router
Dataset Split: full

RouterArena Metrics

Metric	Value
RouterArena Score	0.7612
Accuracy	78.14%
Total Cost	$2.557320
Avg Cost per Query	$0.000304
Avg Cost per 1K Queries	$0.3044
Number of Queries	8400
Abnormal Entries	0
Robustness Score	0.6714

Optimality Metrics

Metric	Value
Opt.Sel (Optimal Selection)	0.1766
Opt.Cost (Cost Efficiency)	0.4549
Opt.Acc (Accuracy vs Optimal)	0.9031

---

Evaluation completed by RouterArena automated workflow

[JiaHg]

@yl231 Please update our results(CrossRouter) to leaderboard. Thanks!

[JiaHg]

@yl231 Please update our latest results(CrossRouter) to leaderboard. Thanks!

[yl231]

Maintainer review — approving. Genuine improvement over the merged Cross-Router (routing pool expanded 3→7 models). Faithfulness verified: sub_10 gap is normal (+5.5%, below the trusted-router average) and Opt.Sel is low (0.18) — i.e. not oracle/label-fit. Official /evaluate: Accuracy 78.14% · Arena 0.7612 · $0.3044 per 1K · Robustness 0.6714. All checks green. Thanks @JiaHg — merging; this supersedes the current Cross-Router leaderboard row.

[yl231]

Heads-up on a leaderboard correction affecting Cross-Router (#172, closes #166) — flagging transparently.

xAI retired the grok-4-1-fast-reasoning slug on 2026-05-15 and silently redirects it to grok-4.3 (stronger, ~5–6× pricier). Cross-Router routed 217 calls to that slug via OpenRouter, and your artifacts confirm they were served grok-4.3 (model_used=x-ai/grok-4.3) — but they had been billed at the retired slug's old $0.20/$0.50 price. We've corrected the accounting to bill the model that actually answered.

Impact: Arena 76.12 → 75.75, cost $0.30 → $0.40/1K — you remain #1. No action needed on your end. Thanks for the submission!