RobustBench-TC - Server 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 an HTTP 5xx error; higher is better.

Metric: Tool-call accuracy (%) under the 5xxErr perturbation (transition component: first call returns HTTP 5xx); 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 April 2028. 18 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Qwen 3 14BNon-reasoning63.3#1269
2O4 Mini61.3#368
3MUA-RL-32B59.8
4MUA-RL-8B53.8#1640
5LoopTool-32B53.3
6Qwen 3 32BNon-reasoning49.7#972
7MUA-RL-14B48.7#1658
8Llama 3.2 3B Instruct45.2#7241
9ToolRL-Llama3.2-3B44.7
10LoopTool-8B44.2#1473
11DeepSeek R1 Distill Qwen 14B42.7#2771
12Qwen 2.5 3B Instruct39.7#5392
13Qwen 2.5 1.5B Instruct38.7#7470
14Qwen 3 8BNon-reasoning37.2#1873
15ToolRL-Qwen2.5-1.5B34.7#2299

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