AGIBOT Worken Proposes a Robot Work Metric

Chinese robot maker AGIBOT wants the industry to stop counting how many robots it builds and start measuring how much work they do. At its Innovation Day in Shanghai on 29 October, it proposed AGIBOT Worken, a metric for “the productive work performed by robots in real-world environments”, the company said. It also launched Link-U OS 1.0, which it calls an operating system for embodied intelligence, and formally launched the G2 Max industrial robot.

Worken, a blend of “Work” and “Token”, is a proposal rather than an agreed standard. AGIBOT calls it “an evolving framework” that it will refine with deployment data.

How AGIBOT Worken is calculated

AGIBOT describes Worken as “Effective Intelligent Work Hours” and gives a simple formula: Worken = Number of Active Robots × Autonomous Working Hours × Output Efficiency Ratio.

Each factor is meant to strip out flattering numbers. The first counts robots actually doing tasks, not units built or sitting in inventory. The second counts time spent working autonomously, not total powered-on time. The third compares output on a defined task against a benchmark, such as the acceptable work a human worker completes under set operating standards.

“We want to move beyond measuring how many robots are built or delivered and begin measuring how effectively they can perform useful work, autonomously and consistently, in the real world,” said Peng Zhihui, AGIBOT’s co-founder, president and CTO. AGIBOT says the metric appears in the “2026 Embodied Intelligence Industry Innovation Report”, compiled by China Economic Information Service and the Shanghai Industrial Innovation Center of the China Academy of Information and Communications Technology.

Link-U OS 1.0 and the open-source promise

AGIBOT calls Link-U OS “the industry’s first operating system purpose-built for embodied intelligence”. It is built around a hardware abstraction layer, an “embodied intelligence-native” kernel and a service framework. AGIBOT says it has “already been validated on mass-produced robots”, is being trialled by selected partners and should become fully open source by the end of 2026. AGIBOT has donated the project to the OpenAtom Foundation, China’s open-source foundation, and joined it as a platinum sponsor.

Some code is already public. A Link-U OS repository on GitHub was created in December 2025 and last received code in January 2026, so the full 1.0 release is still to come.

G2 Max lifts up to 40 kg with two arms

The G2 Max is a dual-arm industrial robot with 21 degrees of freedom. According to AGIBOT’s Chinese launch article, each arm carries 20 kg, both arms together are rated for 40 kg with a 50 kg peak, and repeat positioning accuracy is within ±0.5 mm. It works from 0.1 m to 2.5 m high and has hot-swappable dual batteries for round-the-clock shifts.

AGIBOT claims a success rate above 99% in an e-commerce warehouse, handling 80 boxes an hour when depalletising and 75 when palletising. It says the robot also carries joint parts to final assembly on the line that builds its Qiyuan T1 robot, keeping pace with a target of one T1 every 2.5 minutes. “Embodied intelligence is not a lab demo; its value must be tested on real production lines,” said Yao Maoqing, AGIBOT’s co-founder and co-president (TSN translation). The company also showed new AI models, GO-3 Preview and WITA-Omni, and says GO-3 Preview topped several benchmarks in results presented at the event.

A work metric shifts the humanoid race from shipments to uptime

The humanoid sector has been scored mostly on inputs: robots shipped, funding raised, factories planned. AGIBOT itself has marked unit milestones. AGIBOT Worken is an attempt to change the scoreboard, and its design is sensible. Counting only active robots and only autonomous hours penalises fleets that sit idle or need a human hovering nearby, which is where many pilot deployments struggle.

The hard part is the third factor. An output ratio against a human benchmark depends entirely on who sets the benchmark and for which task. AGIBOT has not published task-specific benchmarks, and no third party audits the inputs. Until that happens, Worken figures will be company claims, as hard to compare across vendors as today’s benchmark leaderboards. The G2 Max numbers above show the problem: a 99% success rate and 80 boxes an hour are useful only if others measure the same way.

That makes the open-source pieces the more consequential news. A shared operating system, if it really opens up, could give other makers a common layer to measure on. Rivals are pursuing the same goal by other routes: NVIDIA’s GR00T and Isaac stack aims to be the default robot software layer, and Unitree has released free swarm-control software for its humanoids. Buyers are already thinking in output terms. KDDI’s ELYZA Robotics talks about 1,000 working robots, and Lotte’s Korean logistics project is built around specific warehouse processes. Even Tesla’s Optimus programme is ultimately judged on factory work.

The test for Worken is simple: whether any company other than AGIBOT publishes a figure using it before Link-U OS goes fully open source at the end of 2026.