TwinRouterBench: Static Agentic LLM Routing Evaluation Set

970 training rows comprise the static track for TwinRouterBench, a benchmark for evaluating realistic agentic LLM routing. The dataset, created by Amorph, was last updated on 2026-05-23. It includes a Hugging Face viewer-friendly table and a separate original question bank.

Use Cases

• Benchmarking static routing performance of agentic LLMs based on the provided question bank.
• Training routing models using the structured training data with nested message fields.
• Comparing routing strategies on a stable schema designed for benchmark consistency.

Strengths

• Contains 970 rows of structured training data.
• Uses a stable schema with JSON strings for nested fields to ensure benchmark consistency.
• Associated with a peer-reviewed paper (arXiv:2605.18859).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full dataset beyond the training split is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Hugging Face dataset uploaded by author 'Amorph'.

Collection Method

Likely created as part of the TwinRouterBench research project for evaluating agentic LLM routing.

Freshness

Last updated 2026-05-23 15:15:22; freshness should be verified.