RobustBench-TC - Duplicate Tool With Swapped Description: 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 a swapped description and wrong parameters; higher is better.

Metric: Tool-call accuracy (%) under the Dup-SwapDP perturbation (action component: same name; swapped 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-32B73
2MUA-RL-14B69#1658
3Qwen 3 32BNon-reasoning64.3#972
4LoopTool-32B64.3
5Qwen 3 14BNon-reasoning61.9#1269
6Qwen 3.5 9BNon-reasoning54.8#828
7O4 Mini52.4#368
8LoopTool-8B51.6#1473
9ToolRL-Qwen2.5-1.5B51.6#2299
10MUA-RL-8B51.6#1640
11Qwen 3 8BNon-reasoning48.4#1873
12DeepSeek R1 Distill Qwen 14B48.4#2771
13TL-CodeLLaMA-246.8
14ToolRL-Llama3.2-3B42.9
15ToolRL-Qwen2.5-3B41.3

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-swapped-description · How It Works · Data refreshed daily, snapshot 2026-11-21.