WorkArena Elo

WorkArena Elo is a ranking score that shows how well an AI model performs real office tasks, based on direct human judgment. Much like how an Elo rating boils down a chess player's skill into a single number, this score comes from having multiple AI models tackle 145 real-world tasks across 29 industries, with human evaluators comparing the results and assigning scores.

A higher number means better skill at office-style collaboration. What sets it apart from other benchmarks that measure coding ability is that it looks at how smoothly a model handles multi-day tasks — like drafting a press release, coordinating schedules, or organizing materials — rather than one-off, single-day jobs.

Articles use it like this: "On the expert-rated collaboration index (WorkArena Elo), Qwen3.8-Max-0902 scored 1468, edging out Claude Opus 5 (1437) but falling short of GPT5.6 Sol (1482)." A common point of confusion is that this score does not represent coding ability. In the same article, coding benchmarks (like Terminal-Bench) and WorkArena Elo are listed as entirely separate categories — a model can lag in coding while still leading in office collaboration.