GuardianAgentBench (Vectara): leaderboard
GuardianAgentBench (Vectara, 2026) runs tool-using agents through 580 business scenarios in six domains, a third of them with adversarially modified tool responses, and requires both the final answer and the tool-call trace to match the ground truth. Every model runs as an agent on the Vectara cloud agent platform. Overall correctness in percent averaged over the six domains, judged by Claude Sonnet 4.5; higher is better.
Metric: Overall correctness (%; mean over six agent domains of the share of scenarios whose answer and tool-call trace are both judged correct by Claude Sonnet 4.5; 580 scenarios with 81 tools, a third with adversarially modified tool responses; Vectara cloud agent platform, default API settings). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Vectara + Claude Opus 4.5 | 74.8 | |
| 2 | Vectara + GPT-5.2 Pro | 72.1 | |
| 3 | Vectara + Gemini 3 Pro | 71.4 | |
| 4 | Vectara + Qwen 3 Max | 68.3 | |
| 5 | Vectara + GPT-OSS-120B | 65.8 | |
| 6 | Vectara + DeepSeek V3.2 | 63 |
Interactive version: theaggregate.ai/benchmark?slug=guardianagentbench-vectara · How It Works · Data refreshed daily, snapshot 2026-11-13.