RobustBench-TC - Rate Limit: 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 a first tool call that returns HTTP 429; higher is better.

Metric: Tool-call accuracy (%) under the RateLim perturbation (transition component: first call returns HTTP 429); 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: Not forecast. 18 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Qwen 3 14BNon-reasoning49.2#1269
2DeepSeek R1 Distill Qwen 14B37.7#2771
3MUA-RL-14B35.2#1658
4Qwen 2.5 1.5B Instruct31.7#7470
5TL-CodeLLaMA-231.2
6LoopTool-8B27.6#1473
7MUA-RL-8B26.1#1640
8ToolRL-Qwen2.5-1.5B25.6#2299
9MUA-RL-32B24.1
10Qwen 3 8BNon-reasoning21.6#1873
11Qwen 3.5 9BNon-reasoning21.6#828
12ToolRL-Llama3.2-3B21.1
13Llama 3.2 3B Instruct20.1#7241
14Qwen 3 32BNon-reasoning20.1#972
15Qwen 2.5 3B Instruct19.1#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-rate-limit · How It Works · Data refreshed daily, snapshot 2026-11-21.