AGIBOT's Worken measures productive robot work in real-world environments. It reports that work as Effective Intelligent Work Hours across a deployed fleet. AGIBOT introduced the framework at AGIBOT Innovation Day 2026 in Shanghai on Oct.
29, according to AGIBOT via PR Newswire in the Innovation Day announcement. An Oct. 30 AI briefing summary says the name blends Work and Token, likened to tokens for language-model output but applied to physical tasks.
Table of Contents
- How does Worken count work?
- Why move beyond units shipped?
- Where do language and action fit?
- What should buyers check before using it?
How does Worken count work?
Worken multiplies three factors: active robots, autonomous working hours, and output efficiency ratio, as reported by Humanoids Daily in the Innovation Day breakdown. The result is a single work-hours figure for a fleet over a period.
Each input has a narrow meaning, according to the Derrington Collaborative AI briefing. The distinction matters because powered-on time and owned units do not prove useful work.
- Active robots: units doing tasks in deployment, not inventory.
- Autonomous hours: self-directed work time, not powered-on time.
- Efficiency ratio: output compared with a task benchmark, such as acceptable human work.
Why move beyond units shipped?
CTO Peng Zhihui framed Worken as a shift away from counting units built or delivered, according to the Derrington Collaborative AI briefing. The message followed AGIBOT's report that its 20,000th robot left production in September 2026.
For readers, the change is from factory scorekeeping to deployment results. A large fleet can still produce little Worken if few units work autonomously or efficiency is low.
Where do language and action fit?
AGIBOT paired Worken with new embodied AI releases, according to RobotHOT. GO-3 Preview handles reasoning, prediction and action, with claimed first-place results on RoboDojo, RoboTwin and RoboColiseum.
The same report describes WITA Omni 1.0 for vision, hearing, language and physical interaction. In practice, spoken or typed commands pass through those models, while Worken measures the resulting productive work.
What should buyers check before using it?
Worken is a vendor proposal, not an agreed standard, according to The AI Insider in its report on the metric. The same report says task benchmarks will be refined from deployment data, so results depend on benchmark choice.
AGIBOT argues that Worken beside lifecycle cost shows rising output from software updates even as hardware depreciates, according to the Derrington Collaborative AI briefing. Treat that comparison as useful only when the inputs are disclosed. Ask for task-level Worken before comparing two deployments.
- Ask how active status is logged per robot.
- Ask how autonomous time excludes idle powered-on time.
- Ask which benchmark sets efficiency for each task.
