The AI arena is free today

Open Superagent
AI OrganizationMaintains 8 benchmarks

XLANG Lab: AI benchmarks & research

University of Hong Kong NLP lab. Maintains OSWorld and its variants.

Maintained by XLANG Lab

8 benchmarks

In one line

XLANG Lab maintains OSWorld (counts toward the LLM Stats Score, 21 models scored), OSWorld 2.0 (counts toward the LLM Stats Score, 17 models scored), OSWorld 2.1 (partial) (counts toward the LLM Stats Score, 11 models scored), OSWorld 2.1 (strict) (counts toward the LLM Stats Score, 8 models scored), OSWorld Extended (counts toward the LLM Stats Score, 1 model scored), OSWorld-G (counts toward the LLM Stats Score, 1 model scored), OSWorld Screenshot-only (counts toward the LLM Stats Score, 1 model scored) and OSWorld-Verified (counts toward the LLM Stats Score, 28 models scored). It ships no models of its own on LLM Stats.

FAQ

Common questions about XLANG Lab.

What is XLANG Lab?

University of Hong Kong NLP lab. Maintains OSWorld and its variants. On LLM Stats it maintains 8 benchmarks, led by OSWorld-Verified, OSWorld and OSWorld 2.0. It ships no models of its own; each score on its benchmarks carries its own provenance label on the model page.

Which XLANG Lab benchmarks count toward the LLM Stats Score?

OSWorld-Verified, OSWorld, OSWorld 2.0, OSWorld 2.1 (partial), OSWorld 2.1 (strict), OSWorld Extended and 2 more count toward the LLM Stats Score. Only benchmarks admitted to the core set move the main leaderboard; the rest are shown for transparency.

Which models lead XLANG Lab's benchmarks?

Qwen3.8 Max leads OSWorld-Verified at 86.1% (28 models scored), Seed 2.1 Pro leads OSWorld at 78.8% (21 models scored) and Claude Opus 5.5 leads OSWorld 2.0 at 81.8% (17 models scored). Each benchmark page lists the full ranking and how the results were produced.

Where are XLANG Lab's official benchmark results?

XLANG Lab's official site is https://www.xlang.ai. Each benchmark page on LLM Stats links to that benchmark's own paper or leaderboard, and shows the scores next to every other benchmark so models can be compared on one page.