Alibaba's AI model tops coding leaderboard, beating OpenAI and Google
Alibaba's new AI model has achieved the highest score in the Code Arena leaderboard, surpassing rivals from OpenAI and Google. The Chinese tech giant is the only non-US company in the top five. Code Arena differs from traditional benchmarks like HumanEval or SWE-bench by testing models on building complete, interactive web applications from scratch based on user prompts. Users vote on anonymized outputs in blind comparisons, reflecting real-world developer preferences. The benchmark is run by Arena, an organization founded by researchers from UC Berkeley, UC San Diego, and Carnegie Mellon University. Alibaba also owns the South China Morning Post.
Key facts
- Alibaba's new AI model scored highest in Code Arena leaderboard.
- It outperformed OpenAI and Google models.
- Alibaba is the only non-US firm in the top five.
- Code Arena tests models on building complete web applications from user prompts.
- Users vote on anonymized outputs in blind comparisons.
- Code Arena is run by Arena, founded by UC Berkeley, UC San Diego, and Carnegie Mellon University.
- Alibaba owns the South China Morning Post.
- Traditional benchmarks like HumanEval or SWE-bench use standardized tests.
Entities
Institutions
- Alibaba
- OpenAI
- Code Arena
- Arena
- University of California, Berkeley
- University of California San Diego
- Carnegie Mellon University
- South China Morning Post
Locations
- China
- United States