List your company

Privacy-Aware Minimal View for Agents on Augmented WebArena N = 662 (test)

96.6Context Recall (TCNP)Llama-3.3-70B-Instruct
94.842495.298795.75596.2113Jun 11, 2026

Evaluation results

MethodLinks
Llama-3.3-70B-InstructProtocol=Zero-shot, Input=Text-only accessibility trees2026.0696.699.631.234.5
GPT-OSS-20BProtocol=Zero-shot, Input=Text-only accessibility trees2026.0696.399.427.632.7
Nemotron-Nano-9BProtocol=Zero-shot, Input=Text-only accessibility trees2026.0695.899.124.929.1
Qwen3-8B-InstructProtocol=Zero-shot, Input=Text-only accessibility trees2026.0695.698.921.128.4
Llama-3-8B-InstructProtocol=Zero-shot, Input=Text-only accessibility trees2026.0695.598.820.627.6
Mistral-7B-v0.3Protocol=Zero-shot, Input=Text-only accessibility trees2026.0695.398.725.730.2
Gemma-3N-E4BProtocol=Zero-shot, Input=Text-only accessibility trees2026.0695.198.519.425.8
MINIMProtocol=Proposed specialized approach, Input=Text-only accessibility trees2026.0694.9199.3110.112