RobustBench-TC - Authentication Error: 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 401 or 403; higher is better.

Metric: Tool-call accuracy (%) under the AuthErr perturbation (transition component: first call returns HTTP 401/403); 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
1TL-CodeLLaMA-235.7
2Qwen 2.5 1.5B Instruct32.2#7470
3DeepSeek R1 Distill Qwen 14B31.2#2771
4ToolRL-Qwen2.5-1.5B30.2#2299
5Llama 3.2 3B Instruct23.6#7241
6Qwen 2.5 3B Instruct19.6#5392
7ToolRL-Llama3.2-3B18.6
8MUA-RL-14B18.1#1658
9Qwen 3 32BNon-reasoning16.1#972
10MUA-RL-8B14.6#1640
11MUA-RL-32B13.6
12Qwen 3 14BNon-reasoning13.1#1269
13LoopTool-32B11.1
14ToolRL-Qwen2.5-3B10.6
15Qwen 3.5 9BNon-reasoning10.1#828

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-authentication-error · How It Works · Data refreshed daily, snapshot 2026-11-21.