ClawWork

Why we included this project

Most benchmarks answer one question: how smart is the model. ClawWork asks whether an agent can actually earn money doing real paid work. Agents work through 220 tasks drawn from actual economic sectors, start with a small cash balance, and are charged for every token they consume. Keep costs low and output quality high and the balance grows; burn tokens without producing usable work and the agent runs dry. The bundled React dashboard tracks each agent's balance, completed tasks, and survival over time, so you can compare model configurations side by side. Teams checking their own LLM setup for cost efficiency, or researchers who want a realistic economic-work evaluation harness, get a runnable framework here rather than a methodology that only exists on paper.