Money closed and models launched, but deliveries and proof did not
Read together, the fortnight points one way: a financing closed and a listing opened, models became more open but stayed vendor-measured, and production claims still ran ahead of delivery evidence. Five conclusions, then a table of what changed.
Agility completed its merger with Churchill Capital Corp XI on Nov 4 and began trading on Nasdaq as AGLT on Nov 5. The company says about $465M gross was raised: about $164M from the trust, about $201M from a Foxconn-led PIPE at $10 per share and $100M of equity received earlier (164 + 201 + 100 = 465). The trust supplied only 35%. Digit 5 orders of more than $300M are conditional on milestones, and RoboFab is designed for up to 10,000 robots a year, which is capacity, not output. Not investment advice.
Gemini Robotics 2.5 (Nov 10) is rolling out to robotics partners only, with 96% success in an Apollo 2 pilot at Mercedes-Benz. NVIDIA's GR00T 2 has a public model page and code "available in December", with 95% success after five demonstrations. AGIBOT's GO-3 Preview claims first place on three benchmarks and promises open source by end of 2026 without a date or licence. No outside party has reproduced any of them.
Agile Robots raised $500M at $8.5B on Nov 12, led by SoftBank with Nvidia taking part. GXO and Apptronik signed only a letter of intent for up to 1,000 Apollo robots over four years, gated by KPIs. Skild AI says it passed $100M in annual recurring revenue. Reported talks, such as Physical Intelligence at about $20B, are unconfirmed and excluded from every total.
Optimus is at "early ramp" with no unit count and V3 has not been shown; XPeng's IRON is targeted for mass production at end of 2026 with deliveries in 2027. The one delivery-like dataset is a third-party estimate: 16,500 humanoids shipped in Q3, of which AgiBot, UBTech and Unitree make up 13,800 or 83.6%.
ASUS showed Gallop on Nov 13 against a 99.5% success target, with customer pilots in 2027 and market availability targeted for 2H 2028. Meta's Max with Fourier is a reported research prototype, Xinhua and UBTech showed bionic news anchors on stage, and Boston Dynamics showed a wheeled Atlas as a research platform. None can be ordered today.
Financings closed and models launched, but delivered units and independent proof of their claims remain 2027 events.
What changed at a glance
Source: Agility releases of Nov 2, 4 and 5; Agile Robots statement via Dealroom and Quartz; GXO release; Google DeepMind; NVIDIA; PR Newswire; ASUS Pressroom; Motor1; Investing.com; Smart Analytics Global.
Prototypes, reports and research videos: nothing new could be ordered
Five items in the window, five evidence types. The only robot with a stated commercial date is ASUS Gallop, and that date is the second half of 2028.
Dated product table
Source: ASUS Pressroom; Humanoids Daily; Unite.AI; Businesskorea; TechNode; Motor1.
Item by item
What was said
Gallop packs 15 small parts in about 20 seconds per board at an ASUS group factory. ASUS's target is 99.5% success within 20 seconds; Unite.AI notes these are targets, not reported results, and Humanoids Daily adds that no measured rate is published.
How it learns and what it runs on
Open AI models, head-camera footage of workers, simulation variations and VR-guided practice; one NVIDIA Jetson AGX Thor and an in-house touch-sensing hand. Fixed base, two arms; wheels come next.
Timing
Customer pilots in 2027, a Computex 2027 showing, market availability targeted for 2H 2028. The CNBC summary describes partner release in 2H 2027 and commercial launch in 2H 2028; this report follows the ASUS blog wording.
Reported details
According to TechNode, Max is 1.5 m and 48 kg, runs on Jetson Thor, has five-fingered hands and was built with Fourier in about three months; Max 2 work began in September.
Status
A research prototype not currently intended for sale. The material contains no Meta statement on security, so none is attributed here.
Wheeled Atlas (Nov 9)
About three minutes of video, a 360-degree spin on the first attempt, built from off-the-shelf wheels and one custom connector. A research platform to test motor and whole-body-control limits. Hyundai Mobis starts prototype Atlas actuators in November; Hyundai targets deployment at its US plant in 2028.
Optimus V3
Still not shown. Musk answered "No" on X on Nov 1 to a question about December consumer orders; Tesla has said nothing officially on a waitlist, fee or price, and the US$20,000 figure is an unconfirmed estimate.
Three releases in one week, three levels of access, no independent proof
Gemini Robotics 2.5, GR00T 2 and GO-3 Preview arrived within about a week. Every headline number is company-reported, and no outsider can yet download or reproduce any of the three.
Model status at a glance
Source: Google DeepMind; NVIDIA research page and blog; PR Newswire; TechTimes; AI Magazine.
GR00T 2 success rates (NVIDIA-reported)
Source: NVIDIA research page. Every bar is an NVIDIA result.
Reading the numbers
Blog versus research page
The NVIDIA blog says GR00T 2 completed 40 of 50 tasks at least once (task coverage, 80% of tasks); the research page's 40% is per-trial zero-shot success on 50 unseen tasks. They are different metrics.
Data mix
Research page: 402,730 hours of egocentric video, 10,184 hours of handheld-gripper demonstrations and 23,500 hours of robot teleoperation, 436,414 in total; the blog's 100,000 hours is the public subset. Egocentric video is about 92.3% of the mix.
Gemini Robotics 2.5 safety
Google reports safe completion up 33% around obstacles and 43% around people, and 99.5% jailbreak resistance for Gemini 4 Argon against 82% for Gemini Robotics ER 2 (a gain of 17.5 points); the 96% Mercedes-Benz result is a single-site pilot.
GO-3 and Skild
GO-3's rankings are self-reported with no date or licence for the promised open source. Skild's over $100M is annual recurring revenue (a run rate), with about $50M recognised.
Lines, capacity and targets are not delivered units
Every figure in the status table is a line, a design capacity, a target or a build count. The only delivery-like numbers are third-party shipment estimates, and Chinese vendors dominate them.
Production and delivery status
Source: New Market Pitch; XPeng; Investing.com; Agility release; Quartz.
Q3 2026 humanoid shipments, units (SAG estimate)
Source: Smart Analytics Global, Nov 8. Last bar is 16,500 minus 13,800 (derived).
Arithmetic and cautions
Shares
AgiBot 4,900 + UBTech 4,500 + Unitree about 4,400 = 13,800, or 83.6% of 16,500; all others together 2,700. A 347% rise implies Q3 2025 at about 3,700 units (derived).
Forecast
SAG forecasts 54,000 units for 2026 and 86,000 for 2027 (estimate). Chinese vendors are about 97% of shipments; full-size bipedal robots reached 47% of units from 38%.
Caution
Shipments are not deliveries to paying industrial customers: Unitree's volume is concentrated in research, education, entertainment and developer use.
Two deals closed; one partnership is only a letter of intent
Closed rows are Agility's merger financing and the Agile Robots Series F. They are unlike events and are not an industry funding total. A report or "in talks" headline is not a round. Not investment advice.
Deal table
Source: Agility releases; Dealroom; Quartz; GXO release; HumanoidMegaHub; optimusk.blog. The closed rows are not additive in kind: 465 + 500 = $965M of gross capital from two unlike events.
Agility gross raise, $M (company release)
Source: Agility pre-close release; components sum to the total, before transaction expenses.
Mechanics and the other deals
How the $465M is built
Churchill XI shareholders approved on Nov 2 with more than 84% of votes cast; the merger completed Nov 4. Components: about $164M trust, about $201M Foxconn-led PIPE at $10 per share, $100M equity already received. The trust supplied 35% (164 / 465).
Lock-up and orders
Existing holders and the sponsor are under a 180-day lock-up from closing, with early release on stock-price measures; counting from Nov 4 puts the end in early May 2027 (derived). Digit 5 orders of more than $300M are subject to contractual milestones; early access is expected in 1H 2027.
Agile Robots
$500M Series F on Nov 12 at $8.5B; total funding about $2B. A factory near Munich can build up to 10,000 humanoids a year (capacity); about 2,000 are targeted for 2026 (Bloomberg, via Quartz). It is a broader industrial-robotics round, not a pure humanoid one.
GXO and Apptronik
A letter of intent, not a definitive contract. From 2027 the trials validate safety and technical requirements against agreed KPIs; subject to success the companies intend to progress toward up to 1,000 Apollo robots with initial North America deployment in late 2027 under a Robotics-as-a-Service model.
Routes differ in hands, data and dependencies, and NVIDIA is the shared layer
Specifications below are company-reported and dated to the cited pages; anything not published is marked "not disclosed". There is no ranking of a "best robot".
Comparison table
Source: Agility; XPeng newsroom; Humanoids Daily; Boston Dynamics; FIRGELLI; ASUS Pressroom; TechNode; Google DeepMind. XPeng's insight page says 22 DoF per hand; this report follows the newsroom's 21.
Degrees of freedom per hand (company-reported)
Source: Boston Dynamics, XPeng newsroom, Google DeepMind. Counts only; no source compares task success.
Who depends on whom
Compute
Gallop, Max, Atlas and Digit 5 use Thor-class NVIDIA computers (four of the five robots that disclose compute); XPeng IRON runs its own Turing chips.
Models
GR00T 2 is the one open model; Gemini Robotics 2.5 is partner-only (Apollo 2 at Mercedes-Benz); GR00T 2 is a world-action model while Gemini Robotics 2.5 is a VLA.
Components
Hyundai Mobis makes prototype Atlas actuators from November; Nyobolt supplies the Digit 5 battery; Fourier built the Max hardware.
Data
GR00T 2 uses 92.3% egocentric human video; ASUS learns from worker head-cameras; 1X argues a video world model needs no large teleoperation set.
What to do: verify the December dates and wait for 2027 deliveries
For every reader the rule is the same: act on delivered units, filed documents and named sites, and treat capacity, targets and benchmarks as claims until a second party confirms them.
Builders and buyers
Agility's orders are conditional, GXO–Apptronik is an LOI with KPIs and Gallop pilots start in 2027. Ask for pilot KPIs, named sites and hours, not capacity.
Investors and strategists
A listing route exists (Agility, AGLT from Nov 5) and Agile Robots was priced at $8.5B; talks are unconfirmed. Compare disclosures, not valuations. Not investment advice.
Model developers
Three release levels: partner-only, open with code due, preview with a promise. Choose by access and licence, and wait for reproducible evaluations.
What would change the reading
An Optimus unit count would turn "early ramp" into output; a named paying IRON customer would test the end-2026 and 2027 plans; an outside re-run of GR00T 2 would show whether 40% zero-shot and 95% few-shot hold. A Gallop miss against 99.5% or a KPI failure in the GXO trials would put 2027 and 2028 volume dates at risk.
Key dates (state as of Nov 17, 2026)
Source: As cited in Sections 1–6. "Promise", "target" and "plan" carry no commitment beyond the company's statement.