RobustBench-TC - Response-Time Description (Neutral Suffix): 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 response-time annotations in the tool descriptions and a neutral suffix on the distractor; higher is better.

Metric: Tool-call accuracy (%) under the TimeDesc-N perturbation (reward component: TimeDesc + neutral suffix); 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
1Qwen 3 32BNon-reasoning60.8#983
2O4 Mini57.8#372
3Qwen 3.5 9BNon-reasoning49#838
4Qwen 3 14BNon-reasoning46.1#1284
5MUA-RL-8B45.1#1656
6LoopTool-8B44.1#1497
7MUA-RL-14B44.1#1674
8MUA-RL-32B43.1
9LoopTool-32B42.2
10DeepSeek R1 Distill Qwen 14B41.2#2739
11Qwen 3 8BNon-reasoning39.2#1890
12ToolRL-Qwen2.5-1.5B37.3#2330
13TL-CodeLLaMA-237.3
14ToolRL-Qwen2.5-3B31.4
15Qwen 2.5 3B Instruct30.4#5410

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=robustbench-tc-response-time-description-neutral-suffix · How It Works · Data refreshed daily, snapshot 2026-11-22.