There is now a humanoid robot success rate in production held over a real shift, in public, and it is a number worth reading closely. In June, AgiBot showed one on a live line. Over a six-day livestreamed run on Longcheer Technology's tablet mass-production line in Nanchang, China, its G2 robots completed 64,828 tasks at a success rate AgiBot puts at 99.99%, over more than 64 hours of continuous operation, and handled 17,625 tablet units, working alongside people (June 29, 2026). This page walks through the run, sets that figure against the bar a buyer draws, and reads what a number from a Chinese line does and does not settle for a plant elsewhere.
What AgiBot streamed at Longcheer
Longcheer Technology assembles tablets at its plant in Nanchang. On its live mass-production line, over six days, AgiBot ran its G2 humanoid robots on three jobs: tablet inspection, sorting out defects, and moving material between stations. The robots worked next to human operators, and the run was carried on a public livestream for the full six days.
AgiBot reports that the G2 robots completed 64,828 tasks over more than 64 hours of continuous operation, handled 17,625 tablet units, and held a success rate of 99.99% across those tasks (June 29, 2026). AgiBot did not state how many robots ran.
The published figures at a glance
| What AgiBot published | Figure | Source |
|---|---|---|
| Tasks completed | 64,828 | AgiBot, June 29, 2026 |
| Success rate across those tasks | 99.99% (AgiBot's figure) | AgiBot, June 29, 2026 |
| Continuous operation | More than 64 hours, over six days | AgiBot, June 29, 2026 |
| Tablet units handled | 17,625 | AgiBot, June 29, 2026 |
The bar a buyer draws
A success rate reads differently to a buyer than to a lab. At the Automate show in June, the director of robotics strategy at a US carmaker said that 80% success is failure in production (Manufacturing Dive, June 30, 2026). That is the bar a run like this is measured against. A demo staged for a minute clears it. A count kept over 64 hours on a live line, streamed for six days so a buyer can watch it, is a harder claim to make and an easier one to check.
The record of who is running humanoid robots and on which lines is growing month by month, and we keep it on a separate page.
Where the line is, and what that settles
The line is in Nanchang, in China, and most operations leaders reading this are not. So it is worth saying what the figure settles and what it does not. It settles that the G2 held a rate on a task in continuous operation, next to people, for days on a real production line. It does not settle how the same robot would run against your parts, your tolerances and your shift, because those are not the ones it ran against.
Four questions to ask of any figure like this
When a maker quotes a success rate for a run on your floor, four questions tie it to your shift.
- What counts as a task, and what counts as a success: a completed move, a correct inspection, a defect caught, or all of them together.
- How long the run held: a rate over 64 hours reads differently from the same rate over one hour.
- What happened at a stop: whether the robot recovered on its own or a person restarted it, and how that was counted.
- How the robot ran when it left the station to charge, since the hours it is away come off the work it completes.
That last question is a shift-planning one as much as a robot one, and we set out the published uptime and charging figures behind it on a separate page.
Reading the run against your own station
Completed work, hours held, first-time success. Those are numbers an operations leader already reads at every station on the floor, and they are the ones a run like AgiBot's is built from. Kept over a shift, they say whether a robot is holding a station or slowing it down.
In a Consultation, we set completed work, hours held and first-time success for one station on your floor, against your own parts and your own shift, so the result sits next to the workflow you run today.
