What it is
Multilingual benchmark for LLM agents with fully localized tasks, tools, and environments across 23 languages and 13 domains. 16,146 agent task instances from 702 canonical tasks across 4 benchmark families, 23 languages, and 13 domains, with prompts, tasks, tool descriptions, and environments all localized.
Key results
- 16,146 task instances from 702 canonical tasks
- 23 languages across 15 scripts and 13 domains
- Apache-2.0; standard release is 5.2 GiB
Useful for benchmarking multilingual tool-using agents on the same tasks across languages.