MindOn Mind-1 robot speed is a company table plus a labeled video

On 30 Sep 2026, MindOn described Mind-1 as a physical-AI model for human-speed manipulation, and that post is the source of the numbers in MindOn Mind-1 robot speed. The company table sets three tasks against a human, Mind-0, and Mind-1. Packaging is 5s human, 10s Mind-0, 3s Mind-1. Parcel sorting is 4s, 6s, and 3s. Folding is 15s, 80s, and 25s. Average inference fell from 82 ms to 32 ms, which the post marks as 61%. End-to-end latency fell from 153 ms to 77.5 ms. Average sensor alignment error is below 5 ms. Those cycle times and those milliseconds are the company post only.

TL;DR

What actually happened

MindOn’s own post is doing the measuring, in the only sense these facts allow. It names Mind-1 as a physical-AI model and pairs it with the phrase human-speed manipulation. The table is the evidence the company offers for that phrase. A second site, the same day, watches a video and refuses to adopt the table as its own lab result.

Read the columns as the company wrote them. Human, then Mind-0, then Mind-1. Packaging: 5 seconds, 10 seconds, 3 seconds. Parcel sorting: 4 seconds, 6 seconds, 3 seconds. Folding: 15 seconds, 80 seconds, 25 seconds. On packaging and parcel sorting, the Mind-1 cell is faster than the human cell in the table. On folding, the Mind-1 cell is 25 seconds against a human cell of 15 seconds, and against a Mind-0 cell of 80 seconds. The table does not say Mind-1 beats the human on every row. Quoting only the two faster rows edits the table.

The latency lines are also the company post. Average inference fell from 82 ms to 32 ms, with 61% printed beside that drop. End-to-end latency fell from 153 ms to 77.5 ms. Average sensor alignment error is below 5 ms. This page does not recompute the percentages or the gaps. The 61% is the company’s figure for the inference drop. The other two lines are stated as the fall and the error bound, without an extra percent added here.

Cycle times and milliseconds stay marked company-only even when a news page links the post. Linking is not re-timing a robot.

The posts are MindOn and Humanoids Daily.

What only the company posted

The seconds and the milliseconds have one author in this story. If a sentence contains 5s, 10s, 3s, 4s, 6s, 15s, 80s, 25s, 82 ms, 32 ms, 61%, 153 ms, 77.5 ms, or below 5 ms, the citation is MindOn’s post. Humanoids Daily did not repeat the cycle times. That absence is part of the record, not a hole this article fills by pretending both sites share the table.

Human-speed manipulation is MindOn’s description of the model. It is the claim the table is supposed to support. Support, here, means “the company published a comparison,” not “an outside clock confirmed the comparison.” MindOn Mind-1 robot speed is easy to headline as a verified win over people. The sourced status is narrower: a company table, plus a video writeup that will not cosign the stopwatch.

Mind-0 appears only as the middle column. The facts used here do not add a product history for Mind-0. It is the prior cell in the same table, slower than Mind-1 on all three tasks as printed: 10s versus 3s, 6s versus 3s, 80s versus 25s. That is a within-company comparison. It is still the company’s hand on the clock.

The 61% sits on the inference line only, from 82 ms to 32 ms. It is not stated as the percent change for end-to-end latency, and this page will not borrow it for the 153 ms to 77.5 ms line. Different measurements, different sentences.

What the second writeup does and does not repeat

Humanoids Daily, on 30 Sep 2026, describes the video. The scenes it names are parcels, shirt folding, and a humanoid collecting laundry. The on-video labels it records are “1.0x Speed” and “Fully Autonomous.” It says those labels, and any speed claim, are company claims, not an independent measurement.

That is a useful kind of writeup. It tells you what the video shows and what the video prints on itself. It does not swap in a new stopwatch. It does not confirm 3 seconds, 25 seconds, or 32 ms. The cycle times were not repeated by Humanoids Daily. If your recap needs those times, you are back on MindOn’s page, and you should say so.

“1.0x Speed” means the video is labeled as not sped up, and that label is a company claim. “Fully Autonomous” is also a company claim. A label burned into a clip is not the same object as a third-party timing study. Humanoids Daily’s point, as used here, is exactly that distinction.

Scenes and tables can drift apart in a retelling. The table’s tasks are packaging, parcel sorting, and folding. The video description adds a humanoid collecting laundry, and it mentions parcels and shirt folding. This page does not force laundry to equal the folding row, and it does not force parcels to equal the sorting row’s 4s / 6s / 3s. Similar chores are not automatic matches to cells the second site never repeated.

There is no contract and no odds market in these facts. A speed claim does not come with a Yes price. Do not invent one so the demo feels more “confirmed.”

What to do as a reader (not a trade)

Watch the video if you want the qualitative look, and read the labels as labels. Then, if you want numbers, open the company table and copy all three tasks, including folding, where Mind-1’s 25s does not beat the human 15s. A thread that only posts packaging at 3s versus 5s has sliced the table.

Keep a bright line on authorship. Cycle times and milliseconds: company post only. Video contents and the two labels: described by Humanoids Daily, and still tagged by that outlet as company claims, not an independent measurement. That second sentence is the correction whenever someone says “a news site confirmed the speed.”

Nothing here is a trade. Mind-1 is a model in a robotics post, not a token call, and this page has no price, no raise, and no “who wins the humanoid race.” Faster cells in a company table are not a profit forecast. They are cells.

Do not extend the 61% beyond inference. Do not turn “below 5 ms” into a guarantee for a robot you have not seen. Do not treat Mind-0’s slower cells as a public benchmark suite. The comparison set is the one the company printed.

Other notes on this site are not timing studies. The DogeOS public testnet does not clock this robot. The Gemini 4 Argon launch does not either. Leave them out of the citation.

A careful reading of MindOn Mind-1 robot speed sounds like this. On 30 Sep 2026 MindOn called Mind-1 a physical-AI model for human-speed manipulation and published the three-row table plus the inference, latency, and alignment figures above. The same day Humanoids Daily described the video and the labels “1.0x Speed” and “Fully Autonomous,” and said those labels and any speed claim are company claims, not an independent measurement. It did not repeat the cycle times. You can be interested in the demo and still refuse to promote the stopwatch by one grade of evidence. Interest is allowed. Upgrading the source is not.

Repeat MindOn Mind-1 robot speed with the folding row still visible. If you only have room for one caution, use the folding row. It stops the slogan “faster than a person” from eating the whole table. Packaging and parcel sorting are the rows where Mind-1’s cell is the quick one. Folding is the row where the human cell is quicker, and Mind-1 is the improvement over Mind-0. All three rows are the company’s. None of them became independent on the day Humanoids Daily described the clip.