RobustBench-TC - Redundant Tool: 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 functionally similar but incorrect tool; higher is better.

Metric: Tool-call accuracy (%) under the RedunTool perturbation (action component: inject functionally similar but incorrect distractor); 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 March 2027. 18 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1LoopTool-32B70.9
2Qwen 3 32BNon-reasoning69.3#972
3Qwen 3 14BNon-reasoning63.3#1269
4MUA-RL-32B62.3
5LoopTool-8B59.3#1473
6ToolRL-Qwen2.5-1.5B59.3#2299
7O4 Mini58.8#368
8Qwen 3 8BNon-reasoning58.8#1873
9Qwen 3.5 9BNon-reasoning58.3#828
10MUA-RL-8B57.8#1640
11ToolRL-Qwen2.5-3B56.8
12TL-CodeLLaMA-256.3
13ToolRL-Llama3.2-3B55.3
14MUA-RL-14B54.8#1658
15Qwen 2.5 3B Instruct53.8#5392

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-redundant-tool · How It Works · Data refreshed daily, snapshot 2026-11-21.