Research finds GPT-4 dominates other LLMs in 'pragmatic tasks of the real world'

The research team behind "Decoder" has developed a benchmark test named AgentBench to measure the capabilities of large language models in assisting tasks. By testing 25 language models, they found that GPT-4 performed the best in both overall scores and across various domains. The team also provides a toolkit, datasets, and a benchmark testing environment for the research community to use. The results of this study are highly valuable for further evaluating the performance of other commercial and open-source models.