ABench: An Evolving Open-Source Benchmark
An evolving open-source benchmark for LLMs on cross-domain tasks
Research paper
Ant Ling introduced ABench, an evolving open-source benchmark suite for evaluating LLMs on complex cross-domain tasks in high-difficulty domains such as physics, actuarial science, logical reasoning, law, and psychology.
The benchmark aims to address evaluation gaps, establish unified standards for multi-domain LLM evaluation, and expand capability boundaries through challenging problems.
