RobustBench-TC - Tool Description Paraphrase: 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 paraphrased tool descriptions; higher is better.

Metric: Tool-call accuracy (%) under the ToolPara perturbation (observation component: tool description); 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
1LoopTool-32B76.4
2Qwen 3 32BNon-reasoning74.9#972
3Qwen 3 14BNon-reasoning72.9#1269
4O4 Mini71.9#368
5Qwen 3.5 9BNon-reasoning69.8#828
6LoopTool-8B69.8#1473
7MUA-RL-8B69.3#1640
8Qwen 3 8BNon-reasoning68.8#1873
9MUA-RL-32B68.8
10ToolRL-Qwen2.5-1.5B68.3#2299
11TL-CodeLLaMA-266.8
12MUA-RL-14B64.3#1658
13ToolRL-Qwen2.5-3B63.8
14Qwen 2.5 3B Instruct62.3#5392
15ToolRL-Llama3.2-3B62.3

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