Benchmark RadarAI BENCHMARK PROFILE
ParamBench
ParamBench is a benchmark for evaluating LLM tool call parameter generation, built from real cloud-network APIs with difficulty tiers and exact match metrics. It is used to evaluate the proposed probe-guided training framework.
- Released
- 2026-08-04
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Tool call parameter correctness is critical for execution yet understudied. ParamBench provides a systematic evaluation of parameter generation across difficulty levels, showing large improvements from probe-guided methods.
Motivation
Large language model agents derive much of their capability from tool use.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.