RobustBench-TC - Response-Time 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 response-time annotations in the tool descriptions and a budget or fast suffix on a distractor; higher is better.

Metric: Tool-call accuracy (%) under the TimeDesc perturbation (reward component: response-time annotation ; same naming pattern); 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-reasoning59.8#972
2O4 Mini56.9#368
3DeepSeek R1 Distill Qwen 14B41.2#2771
4MUA-RL-32B33.3
5MUA-RL-14B33.3#1658
6LoopTool-32B31.4
7Qwen 3.5 9BNon-reasoning30.4#828
8LoopTool-8B30.4#1473
9TL-CodeLLaMA-229.4
10Qwen 3 14BNon-reasoning28.4#1269
11MUA-RL-8B27.5#1640
12Qwen 3 8BNon-reasoning25.5#1873
13ToolRL-Qwen2.5-1.5B24.5#2299
14Qwen 2.5 1.5B Instruct19.6#7470
15ToolRL-Qwen2.5-3B19.6

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