RobustBench-TC - Duplicate Tool With Description and Wrong Parameters: leaderboard

RobustBench-TC (2026) tests tool-calling models under deployment noise grounded in real tool-calling failures, using 199 single-turn samples drawn from BFCL, API-Bank, RoTBench, ToolAlpaca and ToolEyes and scored by deterministic rule-based parsers. This board is tool-call accuracy in percent under an injected same-name distractor tool with the correct description and wrong parameters; higher is better.

Metric: Tool-call accuracy (%) under the Dup-DescWP perturbation (action component: same name; correct description, wrong params); RobustBench-TC samples drawn from BFCL V3 single-turn, API-Bank, RoTBench, ToolAlpaca and ToolEyes, each scored by its source benchmark's strict scorer after a format-tolerant tool-call parse, temperature 0 (DeepSeek-R1-Distill at 0.6), Qwen3-family models with thinking disabled; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 18 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1MUA-RL-32B72
2LoopTool-32B66.4
3Qwen 3 32BNon-reasoning65.6#972
4Qwen 3 14BNon-reasoning57.6#1269
5O4 Mini55.2#368
6LoopTool-8B52.8#1473
7ToolRL-Qwen2.5-1.5B52.8#2299
8Qwen 3.5 9BNon-reasoning51.2#828
9MUA-RL-8B49.6#1640
10ToolRL-Qwen2.5-3B48.8
11Qwen 3 8BNon-reasoning46.4#1873
12MUA-RL-14B46.4#1658
13TL-CodeLLaMA-245.6
14ToolRL-Llama3.2-3B42.4
15DeepSeek R1 Distill Qwen 14B40.8#2771

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=robustbench-tc-duplicate-tool-with-description-and-wrong-parameters · How It Works · Data refreshed daily, snapshot 2026-11-21.