license: mit
language:
- zh
xbench-evals
🌐 Website | 📄 Paper | 🤗 Dataset
Evergreen, contamination-free, real-world, domain-specific AI evaluation framework
xbench is more than just a scoreboard — it's a new evaluation framework with two complementary tracks, designed to measure both the intelligence frontier and real-world utility of AI systems:
- AGI Tracking: Measures core model capabilities like reasoning, tool-use, and memory
- Profession Aligned: A new class of evals grounded in workflows, environments, and business KPIs, co-designed with domain experts
We open source the dataset and evaluation code for two of our AGI Tracking benchmarks: ScienceQA and DeepSearch.
xbench-DeepSearch
DeepSearch is part of xbench's AGI Tracking series, focused on evaluating tool usage capabilities in search and information retrieval scenarios. For detailed evaluation procedures and further information, please refer to the website and Eval Card xbench-DeepSearch (Chinese version)
| Rank | Model | Mode | Company | Accuracy | Evaluation Date |
|---|---|---|---|---|---|
| 1 | o3 | Search | OpenAI | 65+ | 2025.05 |
| 2 | o4-mini-high | Search | OpenAI | 60+ | 2025.05 |
| 3 | Doubao | Deep Think | ByteDance | 50+ | 2025.05 |
| 3 | Grok-3 | DeeperSearch | xAI | 50+ | 2025.05 |
| 3 | Gemini 2.5 Pro (preview) | Deep Research | 50+ | 2025.05 |
Notes
Benchmark data is encrypted to prevent search engine crawling and contamination, please refer to the decrypt code in xbench_evals github repo to get the plain text data. Please don't upload the plain text online.
Submit your agent
If you are developing an AI agent and would like to evaluate it using the latest version of xbench, we welcome you to contact us. Please submit a public access link of your agent, and we will complete the evaluation within an agreed timeframe and share the results with you promptly.
Contact: [email protected]