RobustBench-TC - Malformed Response: 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 malformed JSON; higher is better.

Metric: Tool-call accuracy (%) under the Malform perturbation (transition component: first call returns malformed JSON); 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 November 2027. 18 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1MUA-RL-32B73.4
2MUA-RL-8B67.8#1656
3O4 Mini66.8#372
4LoopTool-8B65.8#1497
5LoopTool-32B60.3
6Qwen 3 14BNon-reasoning59.8#1284
7Qwen 3 32BNon-reasoning59.3#983
8MUA-RL-14B58.8#1674
9Qwen 3 8BNon-reasoning58.3#1890
10ToolRL-Qwen2.5-3B56.8
11Qwen 2.5 3B Instruct54.3#5410
12ToolRL-Llama3.2-3B49.2
13DeepSeek R1 Distill Qwen 14B48.2#2739
14Llama 3.2 3B Instruct47.2#7311
15TL-CodeLLaMA-242.2

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-malformed-response · How It Works · Data refreshed daily, snapshot 2026-11-22.