ToolRL-Qwen2.5-3B: benchmark results
Provider: Other.
Aggregate Rating 1458 ± 52, rank #3438 of 10916 rated models, from 24 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| RobustBench-TC - Schema Drift | 38.2 | Tool-call accuracy (%) under the SchemaD perturbation (trans | 64.7 |
| RobustBench-TC - Duplicate Tool With Wrong Parameters | 52.4 | Tool-call accuracy (%) under the Dup-WrongP perturbation (ac | 52.9 |
| RobustBench-TC - Duplicate Tool With Description | 48.8 | Tool-call accuracy (%) under the Dup-Desc perturbation (acti | 47.1 |
| RobustBench-TC - Duplicate Tool With Description and Wrong Parameters | 48.8 | Tool-call accuracy (%) under the Dup-DescWP perturbation (ac | 47.1 |
| RobustBench-TC - Malformed Response | 56.8 | Tool-call accuracy (%) under the Malform perturbation (trans | 47.1 |
| RobustBench-TC - Query Paraphrase | 64.3 | Tool-call accuracy (%) under the QueryPara perturbation (obs | 47.1 |
| RobustBench-TC - Redundant Tool | 56.8 | Tool-call accuracy (%) under the RedunTool perturbation (act | 41.2 |
| RobustBench-TC - Timeout | 47.7 | Tool-call accuracy (%) under the Timeout perturbation (trans | 41.2 |
| RobustBench-TC - Clean | 63.8 | Tool-call accuracy (%) on the 199 unperturbed baseline sampl | 35.3 |
| RobustBench-TC - Duplicate Tool Without Description | 45.1 | Tool-call accuracy (%) under the Dup-NoDesc perturbation (ac | 32.4 |
| RobustBench-TC - Misleading Description (Neutral Suffix) | 39.2 | Tool-call accuracy (%) under the MisDesc-N perturbation (rew | 29.4 |
| RobustBench-TC - Tool Description Paraphrase | 63.8 | Tool-call accuracy (%) under the ToolPara perturbation (obse | 29.4 |
Interactive version: theaggregate.ai/model?slug=toolrl-qwen2-5-3b · How It Works · Data refreshed daily, snapshot 2026-11-08.