AI
EverydayBench: Evaluating Satisfactory Completion of Live User-Facing Digital Work
Publication: cs.AI updates on arXiv.org
Author: Fangrui Huang, Bingyang Wang, Chuhan Li, Yijiang Li, Arpandeep Khatua, Austin Shen, Riki R Z Choi, Cuiyuanxiu Chen, Jin Huang, Zachary Robertson, Kate Lin, Akshay Paruchuri, Sijun Tan, Xin Liu, Zhengm
Source published: 20/11/2026, 05:00:00 UTC
Briefing created: 20/11/2026, 05:32:50 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 20/11/2026, 05:03:24 UTC.
Briefings
EverydayBench introduces a benchmark of 77 human-authored tasks evaluating AI agents' ability to complete live user-facing digital work. The system combines deterministic constraint checks, real-time web verification, and rubric-based quality assessment. Across 23 agent-model configurations, Claude Opus 5.5 and Claude Fable 5.1 achieved the highest strict pass rate at 54.55%, while GPT-5.6 Sol led lookups and Fable 5.1 excelled in planning. Reliability varied significantly by task type, highlighting the difficulty of delivering complete, well-supported outcomes for everyday requests.
Key points
- EverydayBench introduces 77 human-authored tasks evaluating AI agents' ability to complete live user-facing digital work.
- Claude Opus 5.5 and Claude Fable 5.1 achieved the highest strict pass rate at 54.55% across 23 agent-model configurations.
Context
This benchmark addresses the gap in existing evaluations by integrating predefined success criteria, current evidence verification, and deliverable-quality assessment for AI agents selecting outcomes on the live web.
Limitations
Evaluation relies on human judgments and controlled grounding failures; reliability varies substantially by task type. Preprint status (arXiv:2611.25146v1) and snapshot limitations apply to the evidence.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 20/11/2026, 05:32:50 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeif6w563i7enftx245kxnlgdxztag6r2oi3glofaeysygyukuhce6q