CodeRouterBench: Benchmark Results for Agentic Model Routing on Coding Tasks

CodeRouterBench is the benchmark data released with the Agent-as-a-Router research project. The core unit is a complete task-by-model result matrix where every benchmark task has one recorded result for each of eight canonical backend models. It was created by Lance1573 and last updated on Hugging Face in June 2026.

Use Cases

• Evaluating the performance of different backend models on coding tasks based on the task-by-model result matrix.
• Training or fine-tuning router models for AI agents based on the benchmark results.
• Comparing model routing strategies for code generation based on the provided task results.
• Analyzing the consistency of model performance across a set of coding tasks.

Strengths

• Provides a complete task-by-model result matrix for eight canonical backend models.
• Associated with a published research paper and open-source repository, providing context.
• Includes an optional trained router adapter (Lance1573/acrouter-qwen35-08b-router-lora) for practical application.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and dataset size are unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Released with the Agent-as-a-Router project (GitHub: LanceZPF/agent-as-a-router).

Collection Method

Likely generated from evaluating eight backend models on a set of benchmark coding tasks.

Freshness

Last updated 2026-06-23 17:53:28; freshness should be verified.