Humanoid robots and robotic hands: a 2026 global price guide

Blog · Robotics manufacturing Humanoid robots and robotic hands: a 2026 global price guideFrom a $1,400 developer kit to a $300,000+ Atlas, here is the full 2026 map of humanoid robots, dexterous hands, and arms — who makes them, what they cost, and which tier fits your build. This is the companion to our earlier read on whether humanoids will replace the robot vacuum. That piece argued that Unitree’s ¥9,900 seven-axis arm is a signal, not a finished product. Here we zoom out and price the whole field, so a hardware founder or procurement lead can see exactly what each dollar buys and where the floor is heading. The 2026 price spectrum at a glance The table below lists the major announced or shipping affordable, developer, and research robots and robotic hands, with country, maker, release window, launch price in US dollars (approximate, converted from local list prices where noted), and the technical headline that matters to a buyer. RMB prices are converted at about 7.15 CNY/USD. Read it once for the floor — how low a humanoid can go — and again for the spread, because the gap between a $1,400 kit and a $300,000 platform is not noise: it is the difference between a benchtop toy and a torque-controlled research machine. One caveat before you read any price column: a launch price is a snapshot, not a guarantee. Several makers have already cut list prices between announcement and today, and the direction of travel in 2026 is unambiguously down. Figures marked “est.” or “target” are estimates or stated goals, not list prices — conflating the two is how procurement plans go wrong. CountryMaker & modelReleasedLaunch price (USD, approx)Technical highlightAvailability ChinaUnitree R1Jul 2025~$5,600 (¥39,900)26 DOF, 25 kg, runs and cartwheels, on-board multimodal LLMShipping ChinaUnitree G12024~$16,000 US list (¥99,000 in China)23–43 DOF (base–EDU), 127 cm, Jetson Orin, ROS 2Shipping ChinaUnitree H12023~$90,000180 cm, 26 DOF, high-torque actuatorsShipping ChinaUnitree H22025$29,900Research and education tierShipping ChinaUnitree 7-axis bionic armAug 2026~$1,380 (¥9,900)7-axis joint, 5.5 kg, 650 mm reach, plug-and-playNew release ChinaUBTech Walker S12024$40,000–60,00041 DOF, 172 cmShipping ChinaFourier GR-22024$100,000–150,000 (est.)53 DOF, 175 cmShipping ChinaAgibot A2 (Zhiyuan)2025$50,000–80,00040+ DOF, interactive servicePre-order ChinaLinkerHand L10 / L20 (Lingxin)2025~$2,800 / ~$7,000 (¥19,999 / ¥49,999)20+ DOF dexterous handShipping ChinaZhiyuan OmniHand 2025Aug 2025~$2,070 (¥14,800)16 / 19 DOF hand, tactile optionShipping ChinaUnitree Dex5 handApr 2025~$2,000–2,800 (¥14k / ¥20k)20 DOF, 94 tactile pointsShipping ChinaNoetix Bumi2025$1,400Cheapest full humanoid kitShipping Norway / US1X NEOOct 2025$20,000 or $499/mo168 cm, tendon-driven, dexterous hands, home usePre-order (2026 US) USATesla OptimusTarget$20,000–30,000 (target)28+ DOF, no external sales yetPrototype USAFigure 02 / 03Pilot~$130,000–150,000 (est.)Enterprise, BMW pilotsPilot USAApptronik Apollo2025~$50,000 (target under $50k)71 DOF, 22 h battery, Mercedes / Jabil / GXODeploying USABoston Dynamics Atlas (electric)2024$150,000–320,000 (est.)electric, Hyundai-ownedCommercial pilots USAAgility Digit2020+~$250,000, RaaS leasingWarehouse, Amazon deploymentsCommercial UKShadow Robot Hand—>$100,000Research-grade 5-finger handShipping SpainPAL Robotics TALOS—$300,000+32 DOF, torque-controlledShipping France / JapanAldebaran NAO / Pepper (SoftBank)—$8,000–12,000 / $25,000–30,000Education / service humanoidShipping Which robot fits your budget and use case For a sourcing engineer. A price map is only useful if it maps to what you are trying to do. Four budget bands cover most buying decisions — pick by use case first, then by price. Here is the short version, and below it what each tier actually buys you in 2026. Budget bandRepresentative models (launch price)Who it suits Under $3,000Noetix Bumi $1,400 · Unitree 7-axis arm ¥9,900 · LinkerHand L10 ¥19,999 · Zhiyuan OmniHand ¥14,800Grasping and manipulation algorithm work, embodied-AI demos, lab teaching $3,000–20,000Unitree R1 $5,600 · Unitree G1 $16,000 · UBTech Cruzr $15k–20k · Aldebaran NAO $8k–12kHumanoid locomotion and manipulation research, development platforms, service robots $20,000–100,0001X NEO $20k · UBTech Walker S1 $40k–60k · Fourier GR-2 $60k–80k · Agibot A2 $50k–80k · Apptronik Apollo ~$50kFactory pilots, light manufacturing, commercial service trials $100,000+Unitree H1 ~$90k · Figure 02 ~$130k–150k · Atlas $150k–320k · Digit ~$250k · TALOS $300k+ · Shadow Hand >$100kHeavy research, enterprise robotics-as-a-service, high-end dexterous hands Tier 1 — Developer kits and components, under $3,000 The Noetix Bumi at $1,400 is the cheapest full humanoid kit money can buy — a benchtop platform for teaching and demos, not a machine for a paying customer. The real signal here is the Unitree 7-axis bionic arm at ¥9,900 (about $1,380): the company’s first sub-¥10,000 manipulator. Pair it with a LinkerHand L10 (¥19,999) or a Zhiyuan OmniHand (¥14,800) and you have a manipulation research cell for under $5,000. This band suits universities, embodied-AI startups doing grasping and policy-learning work, and hardware teams validating an actuator or perception stack. What you are not getting is heavy payload, industrial duty cycles, or round-the-clock reliability — treat everything here as a consumable learning tool and budget for spares. Tier 2 — Research and development platforms, $3,000–20,000 This is where a full, buyable humanoid becomes realistic. The Unitree G1 at about $16,000 is the default: a 127 cm platform with 23 to 43 degrees of freedom depending on configuration, a Jetson Orin on board, and ROS 2 support, so your existing software ports over without a rewrite. The R1 (from ¥39,900, roughly $5,600) is the budget sibling aimed at locomotion and embodied-AI research. Aldebaran’s NAO ($8,000–12,000) remains the education stalwart on maturity, not specs. This band is for labs and teams building proof-of-concept before spending six figures. The trade-off is durability and support depth: these machines are built to be programmed against, not to run three shifts unattended. Tier 3 — Industrial pilots, $20,000–100,000 From here up you are buying a machine that could plausibly do paid work. 1X NEO enters at $20,000 (or $499 a month) targeting the home. UBTech’s Walker S1 ($40,000–60,000), Fourier’s GR-2 ($100,000–150,000 (est.)), and Agibot’s A2 ($50,000–80,000) are the Chinese factory-pilot cohort, while Apptronik’s Apollo targets under $50,000 in pilots with Mercedes, Jabil, and GXO. At this tier the decision stops being about sticker price and starts being about uptime, safety certification, and integration cost. A $60,000 robot that needs $40,000 of tooling and six months of integration is a different proposition from one that drops into an existing cell — ask for the integration cost up front, because it is the line item most buyers forget. Tier 4 — Enterprise and research flagship, $100,000+ The top of the market is heavy research platforms, robotics-as-a-service fleets, and Western machines not yet on open sale. Unitree’s H1 (~$90,000) and PAL Robotics’ TALOS ($300,000+) are torque-controlled research flagships. Boston Dynamics’ electric Atlas (est. $150,000–320,000) and Agility’s Digit (~$250,000) are commercial platforms sold through pilots or RaaS leases. Figure’s 02/03 (est. $130,000–150,000) and Tesla’s Optimus ($20,000–30,000 target) remain pilot-stage or pre-sale. The price here reflects torque density, reliability engineering, and a human-safety envelope the budget tier lacks. If your use case demands certified human-robot collaboration, this is where the compliance is built in. What you can actually buy today Three tiers are clear in 2026. The budget tier is led by Unitree’s R1 at about $5,600 and the Noetix Bumi at $1,400, both aimed at developers and universities rather than factories. The mainstream buyable humanoid is the Unitree G1 at roughly $16,000, with UBTech, Fourier, and Agibot shipping in China at $40,000–80,000. At the top, 1X NEO opens pre-orders at $20,000 (or a $499 monthly subscription) for home use, with first US deliveries in 2026. What you cannot buy yet matters as much as what you can. Tesla Optimus remains a prototype with no external sales channel; Figure runs enterprise pilots; Atlas and Digit are available through commercial pilots or robotics-as-a-service leases rather than outright purchase. For a team that needs a robot on the bench this quarter, the realistic shortlist is Unitree, UBTech, Fourier, Agibot, and (soon) 1X. The sub-$6,000 segment has already moved within 2026. Unitree’s R1 line now spans an R1 AIR variant at ¥29,900 (about $4,200) and the standard R1 at ¥39,900, and the company shifted the standard model to in-stock retail around mid-2026 (per Unitree’s product page and Chinese tech-press coverage). For a buyer, that changes the calculus: a robot you can get this month for a pilot beats a cheaper robot you might get in two quarters. Dexterous hands and arms: the “last centimeter” For a hardware architect. Hands are the most expensive single component of a humanoid, often 20–30% of the bill of materials, and the hardest to mass-produce. The range is extreme. Western research hands like the Shadow Robot Hand exceed $100,000. Chinese makers have pulled the floor down hard: Unitree’s Dex5 hand (20 DOF, 94 tactile points) lands near $2,000–2,800, LinkerHand’s L10 at about $2,800, and Zhiyuan’s OmniHand 2025 at roughly $2,070. Unitree’s bionic 7-axis arm, launched 20 August 2026 at ¥9,900 (about $1,380), is the newest data point: the company’s first sub-¥10,000 manipulator. At 5.5 kg with a 650 mm reach and plug-and-play setup, it targets labs, educators, and lightweight pick-and-place rather than heavy industry, and it can pair into dual-arm sorting and assembly cells. For a buyer, that price point turns a robotic arm from a capital-approval item into an everyday line item. Buyer’s noteThe exact torque of Unitree’s new 7-axis arm was not fully disclosed at launch. Treat the ¥9,900 figure as an entry price for research, education, and light tasks — confirm the specification sheet before any production-adjacent use. What the launch coverage did disclose, across Chinese tech press, fills in the buyer-relevant picture: a 7-axis design with a stated 2 kg rated payload, roughly 0.1 mm repeatability, a 24–60 V supply, XT30 power and CAN485 control interfaces, and a Ubuntu-based control stack with position and force-control modes plus collision detection. Unitree has said it will progressively open-source the control program, with a quick-change end-effector mount that swaps between two-finger grippers, three-finger actuators, and five-finger hands. The arm is priced as a platform — the hand is where the rest of your budget goes. How a dexterous hand is driven, and why it costs what it does If you are sourcing hands rather than buying a whole robot, three drivers explain most of the price before you compare two datasheets that look identical. Degrees of freedom (DOF). Every independently actuated joint adds a motor, a transmission, and a control channel. A two-finger gripper has one to three DOF; a five-finger hand with an opposable thumb and lateral finger spread runs 12 to 21. More DOF means more objects it can grasp, but more points of failure. For most pick-and-place, a 3-finger force-controlled gripper beats a 20-DOF hand on reliability and cost; the 20-DOF hand only earns its keep when the task is genuinely dexterous. Actuation and transmission. Expensive hands use direct-drive or quasi-direct-drive motors at each joint, or tendon drives that route the motors into the forearm to keep the fingers slim. Budget hands use smaller hollow-cup or coreless motors with gear or linkage transmissions, trading fingertip force and compliance for cost. The transmission is where the reliability gamble lives: a tendon that stretches or a linkage with play shows up as grasp drift — dropped parts in a production cell. Sensing. This is the fastest-moving cost line. A basic hand has joint encoders; a good one adds fingertip force sensing; a research-grade hand adds dense tactile arrays — Unitree’s Dex5 advertises 94 tactile points — plus slip detection. Tactile silicon, wiring density, and calibration are a meaningful fraction of the hand’s price and the part most likely to fail in the field. If your application only needs to know “am I gripping it,” do not pay for “how hard, exactly where, and is it slipping.” The price gap you should understand before you order The price of a given hand depends heavily on where you buy it. The same class of hand that lists at ¥14,000–20,000 ($2,000–2,800) from a Chinese maker can appear on a US retail storefront at several times that: independent August 2026 pricing puts a Unitree Dex5-class hand at roughly $24,000–27,000, an AGIBOT OmniHand at $5,999–19,999 depending on configuration, and a BrainCo Revo 3 Vision at up to $35,000 (sources: GrabaRobot and US Robot Store). The gap is retail markup, import logistics, and warranty burden — not a different product. That spread matters for two reasons. First, if you are sourcing hands at volume to manufacture, quote the component from the maker or an integrated contract manufacturer, not a Western storefront — the retail price is a ceiling, not your cost basis. Second, it explains why teams move their manipulator bill of materials to a Pearl River Delta supplier once volumes get real. Analysts peg the hands on a Tesla Optimus-class machine at roughly $9,500 of a ~$55,000 build — about 17% — which is why hands are the first place cost engineering goes to work. Why prices are collapsing The drop is fast enough to plot by year: 2024 — Unitree G1 at $16,000 becomes the first full-size humanoid to break $20,000. Jul 2025 — Unitree R1 at ¥39,900 (~$5,600) takes a full humanoid under $6,000. Aug 2026 — the ¥9,900 (~$1,380) 7-axis arm puts a bionic arm under ¥10,000, pointing squarely at the $1,000 hand. Three forces explain the divergence between Chinese and Western pricing. First, vertical integration: Unitree builds its own actuators, drawing on nearly a decade of quadruped volume. Second, a dense local component ecosystem — hollow-cup motors, harmonic reducers, and force sensors are now largely domestic, with localization rates cited at 44–70% depending on the part. Third, the manufacturing base itself: the same Pearl River Delta supply chain that produces consumer electronics can absorb robotics volume. That last point is the one most relevant to a hardware buyer. The robot you eventually ship is won or lost at the bill-of-materials stage — PCBA yield, CNC-machined structural tolerances, harness reliability, and assembly consistency. A $5,000 robot is only $5,000 because someone engineered the cost out of every subassembly. The people running these companies are telling you the same direction. At the 2026 World Robot Conference, Unitree founder Wang Xingxing framed the consumer timeline as “two to three years if fast, five to ten if slow” for a robot that does chores, with the tipping point at roughly 80% of everyday tasks from a voice command in an unfamiliar home. That is a software milestone, not a hardware one. A buyer’s decision framework: use case × budget × lead time For a program manager. Strip away the hype and a humanoid purchase reduces to three questions, answered in this order. 1. What is the task, and does it need a humanoid at all? Most manipulation work in a factory does not need legs, and a lot of it does not need a five-finger hand. If your task is “pick this box from a fixed pose and place it,” a $1,380 arm with a two-finger gripper is the right spend, not a $60,000 humanoid. The humanoid form factor earns its keep only where the workspace was built for humans and cannot be rearranged. Write the task down before you write the purchase order. 2. What budget band does the task actually require? Use the four tiers above as a sanity check. Research and demos belong under $3,000; a proof-of-concept sits in $3,000–20,000; a factory pilot needs $20,000–100,000; only certified, round-the-clock, or high-torque work justifies six figures. The most common mistake is over-buying a flagship for a task a mid-tier platform handles — or buying a developer kit and expecting production uptime. 3. When do you need it, and what is the real lead time? In 2026 some models are in-stock retail (Unitree’s R1 line), some ship in weeks to months (G1, H1, Walker S1), some are pre-order with a stated window (1X NEO, 2026 US), and some have no open sales channel at all (Optimus, Figure, Atlas, Digit). If your pilot is gated on a robot that cannot be bought, your pilot is gated on a partnership negotiation. A university lab validating a grasping algorithm should buy the ¥9,900 arm and an OmniHand in stock this month; a startup proving a warehouse task should buy a G1 and budget for integration; an enterprise scaling a logistics pilot should weigh RaaS leases against purchase. Match the tier to the task, confirm the lead time, and only then compare vendors on price. What this means for robotics hardware buyers If you are building a robot, the price map says two things. One, do not over-invest in in-house manufacturing of parts you can source or contract more cheaply — the leaders win on supply chain, not on doing everything themselves. Two, the differentiators that survive price compression are integration quality, reliability, and time-to-volume, all of which depend on a manufacturing partner who can take you from prototype to mass production without a second source. For teams at the EVT-to-PVT stage, the practical move is to lock the design, then contract the electronics and machined structure to a partner with EMS and CNC under one roof. That removes a coordination layer and keeps the iteration loop tight when specs change weekly. The same logic applies if you are buying rather than building. When you evaluate a maker, ask three questions: does it self-develop and trace its core actuators, or assemble someone else’s parts? Does its manufacturing partner hold the certifications — ISO 9001 and IATF 16949 for the electronics, plus documented tolerance control for the machined structure — that predict consistent units? And how does it handle the iteration that follows your first field test? A supplier who can re-spin a PCBA and re-machine a bracket in the same week is worth more than a 5% discount from one who cannot. Keep reading Will Humanoid Robots and Dexterous Arms Replace the Robot Vacuum? Robotics EMS in China: Drivers, Sensors and Control Boards Robotics CNC Machining in China: Joints, Housings and Tolerances Contract Manufacturing in China: A Buyer’s Guide FAQ What is the cheapest humanoid robot available in 2026?The Noetix Bumi is the cheapest full humanoid kit at about $1,400. Among full-featured, buyable humanoids, Unitree’s R1 is the lowest at roughly $4,200 (¥29,900 at 2026 pricing). For manipulation rather than a whole body, Unitree’s new 7-axis bionic arm at ¥9,900 (about $1,380) is among the cheapest dexterous bionic arms announced so far, and the first sub-¥10,000 manipulator from a major humanoid maker. Education bots like Aldebaran’s NAO start near $8,000. Which humanoid robots can I actually buy and deploy today?Unitree (G1, R1, H1, H2), UBTech, Fourier, and Agibot are shipping, with Agibot and some others in pre-order. 1X NEO is taking pre-orders for 2026 US delivery. Tesla Optimus, Figure, Atlas, and Digit are not generally purchasable — they are in prototype, enterprise-pilot, or lease-only stages. How much does a dexterous robotic hand cost?From about $1,400 for budget Chinese hands to more than $100,000 for Western research hands like Shadow. Mid-market Chinese hands — Unitree Dex5, LinkerHand L10, Zhiyuan OmniHand — sit in the $2,000–7,000 range at the maker, though the same hands can list for $6,000–27,000 on Western retail storefronts. Hands typically represent 20–30% of a humanoid’s bill of materials. Why are Chinese humanoid robots so much cheaper?Vertical integration of actuators, a mature local component ecosystem (motors, reducers, sensors now largely domestic), and the Pearl River Delta manufacturing base that already produces consumer electronics at scale. The cost advantage is structural, not a discount. Are Chinese humanoid robots reliable enough for production use?For research, education, and pilot use, platforms from Unitree, UBTech, Fourier, and Agibot are proven and widely deployed. For round-the-clock production, be honest about the real bottleneck: component life — hollow-cup motor life is often quoted in the low thousands of hours — still limits 24/7 industrial duty everywhere, not just in China. When evaluating any platform, check three things: whether the maker self-develops and can trace its core components (Unitree reports over 90% in-house), whether it works with a certified volume manufacturer (ISO 9001 / IATF 16949), and how it handles after-sales and iteration. This is why robotics teams increasingly pick an EMS + CNC partner with those certifications, sited in the same region as the component supply chain. What is the lead time on a humanoid robot in 2026?It spans the whole field. Some budget models are in-stock retail — Unitree moved its R1 line to in-stock availability around mid-2026 — while mainstream platforms like the G1 and H1 ship in weeks to months. Pre-order programs like 1X NEO carry a stated delivery window (2026 for the US). Prototypes such as Optimus, Figure, Atlas, and Digit have no open purchase channel. Confirm current availability before you plan a program around a specific robot — the gap between “announced” and “on your loading dock” is where schedules die. Should I buy a whole robot or just source the components and build my own?If your task maps cleanly onto an existing platform, buy the robot and spend your engineering effort on the application. If your differentiation is a custom end-effector, actuator, or form factor, source the components and build. The tipping point is volume and uniqueness: below a few dozen units buying wins; beyond it, building with a contract manufacturer wins. Most teams land in between — they buy the base platform and contract the custom parts, where an EMS + CNC partner with no minimum order quantity earns its keep. On the floor the checks are concrete rather than aspirational. Every BOM line is cross-checked against the drawing and the purchase spec before release, so a material callout cannot silently change grade, and critical alloys and components are bought through a cross-checked second source with a matching mill test certificate. For aerospace programs we run FAI per AS9102 so the first article is dimensionally signed off before the run begins, under an AS9100-aligned quality system. Building a robot — or buying one to integrate?Nex-G does EMS and CNC machining in Dongguan, China — No MOQ, from EVT to mass production, under ISO 9001 / IATF 16949 / ISO 14001. One supplier for the board, the actuator housing, and the harness.Get a quote Sources & coverage: prices and dates compiled August 2026 from manufacturer pages and industry trackers; figures marked “est.” or “target” are estimates, not list prices. New data verified via WebSearch in August 2026: Unitree 7-axis arm R1 specs (2 kg rated payload, ~0.1 mm repeatability, 24–60 V, XT30 + CAN485, Ubuntu control stack, position + force-control modes) from Phoenix Tech / Sina Tech / Science and Technology Innovation Board Daily (20 Aug 2026); Unitree R1 humanoid line-up and in-stock shift (R1 AIR ¥29,900, R1 ¥39,900) from unitree.com; dexterous-hand retail pricing spread (Dex5-class ~$24,000–27,000, OmniHand $5,999–19,999, BrainCo Revo 3 up to $35,000, Morgan Stanley 17% hands estimate) from GrabaRobot and US Robot Store (Aug 2026); Wang Xingxing’s WRC 2026 remarks. Third-party retail prices reflect storefront listing prices and differ from maker domestic list prices. Further reading: Unitree · The Robot Report · Interesting Engineering · RoboAtlas · Jiemian News. RMB prices converted at about 7.15 CNY/USD. This is editorial analysis, not a live quotation — request a formal quote for pricing and lead time.