[Hyeon-kki]

Summary

Submit ATRouter, a closed-source router.

• Full predictions: 8,400
• Robustness predictions: 420
• Submitter: @Hyeon-kki
• Local format and pricing checks passed.
• Total cost: approximately $0.6082 per 1,000 queries, including routing calls.

Provenance

• ATRouter uses fixed profiles calibrated on a separate development collection. This collection consists of 9,736 items, after excluding samples that exactly matched or were semantically similar to RouterArena inputs.
• Training was done without common instruction prompts. Routing decisions use query content without accessing reference answers or per-model correctness labels.

Notes

• MMLUPro_health_4857 received no answer because Gemini 3.8 Flash blocked the request with a SAFETY error. It remains recorded as a failed generation in the submission.

[Hyeon-kki]

/evaluate

[github-actions]

Router Evaluation Results

Router: ATRouter
Dataset Split: full
Evaluator: main @ 44cc1ff

RouterArena Metrics

Metric	Value
RouterArena Score	0.7630
Accuracy	79.58%
Total Cost	$5.108653
Avg Cost per Query	$0.000608
Avg Cost per 1K Queries	$0.6082
Number of Queries	8400
Abnormal Entries	1
Robustness Score	0.8548

⚠️ 1 of 8400 queries (0.0%) had no valid generation (inference failed / empty answer) and were scored as incorrect (0). These queries still count toward the denominator, so accuracy and cost reflect the full query set. Please regenerate predictions for these queries and resubmit for a complete evaluation.

---

Evaluation completed by RouterArena automated workflow