Alibaba's 27B-Parameter Qwen Model Rivals OpenAI's GPT-5.6 Luna in Benchmark Tests
Last Friday, Alibaba unveiled its Qwen3.8-27B, boasting 27 billion parameters, which, as per Artificial Analysis, performs on par with OpenAI's GPT-5.6 Luna. On Monday, the benchmarking firm released test results demonstrating that the output quality of Qwen3.8-27B matches that of GPT-5.6 Luna, introduced by OpenAI the previous month. Furthermore, Qwen3.8-27B surpassed GPT-5.6 Terra and Anthropic's Claude Opus 4.8 on the Agentic Index, which assesses AI agent workflows. These findings suggest that Alibaba's model stands strong against larger open-source alternatives. By making the model weights available, Alibaba allows developers to validate benchmarks and incorporate the model, differing from OpenAI's restricted API strategy. This development occurs amid intensifying competition among AI laboratories prioritizing efficiency and transparency.
Key facts
- Alibaba released Qwen3.8-27B model weights last Friday.
- Artificial Analysis published benchmark results on Monday.
- Qwen3.8-27B has 27 billion parameters.
- It performed on par with OpenAI's GPT-5.6 Luna.
- GPT-5.6 Luna is the most cost-efficient model in OpenAI's GPT-5.6 family.
- OpenAI launched the GPT-5.6 family last month.
- Qwen3.8-27B outperformed GPT-5.6 Terra and Anthropic's Claude Opus 4.8 on the Agentic Index.
- It nearly matched DeepSeek and Zhipu's open-weight models.
Entities
Institutions
- Alibaba
- OpenAI
- DeepSeek
- Zhipu
- Anthropic
- Artificial Analysis
Locations
- China
- United States