Hugging Face Ships $2,500 Open Humanoid Robot With 7B Model

Hugging Face released the LeRobot Open Humanoid v0.1 on June 10, 2026. The standard kit sells for $2,500. The package includes a 6-DoF upper body, two 6-DoF arms, a differential drive base, an RGB-D camera, and two-finger grippers. The base kit runs a 3 billion parameter vision-language-action model locally. A 7 billion parameter policy model ships with the Pro configuration. Both models use an 8,192 token context window. The software stack carries an Apache 2.0 license. Hardware design files use the CERN OHL-P license. You can read the official announcement on the Hugging Face homepage. Builders who want open model options can check the best free AI models list for related model releases.

The LeRobot team at Hugging Face built the platform in public. The team shared the release on the vendor homepage and on GitHub. No private repository is required. Users can download assembly guides, PCB files, CAD models, and firmware from the public sources. The robot stands 1.2 meters tall and weighs 9 kilograms in the base version. It uses a Raspberry Pi 5 with 8 GB of RAM. The base kit runs Ubuntu 24.04 and ROS 2 Jazzy. The Pro kit swaps in an NVIDIA Jetson Orin NX for more local compute. The project grew out of the LeRobot open source library for robot learning. The goal is to lower the cost of real world robot data collection.

The $2,500 price matters because closed humanoid platforms remain expensive. The Unitree G1 starts around $16,000. Figure and Tesla humanoids are not sold as open research kits. Hugging Face wants a repairable platform that schools and small labs can afford. Every motor, bracket, and board can be replaced. The license allows commercial use without royalties. The initial manipulation tests reported a 72 percent success rate on pick-and-place tasks for the 7B model in unseen environments. That is below industrial robot reliability, but it is enough for research. The release also shows how local models can bypass cloud API fees. AI free tier limits have pushed more developers toward local and open inference.

Pre-orders opened on June 10, 2026. The first production batch is limited to 500 units. Shipping starts in July 2026. The launch includes a free simulation environment with MuJoCo and Isaac Sim support. Teleoperation tools for the physical arms are included in the research bundle. The project uses a rolling release model. Hugging Face plans monthly software updates and quarterly hardware revisions. A public roadmap tracks new gripper options and battery upgrades. This release is not a one-off demo. It is an attempt to build a low cost open humanoid ecosystem around the LeRobot stack. Early orders will receive assembly support through the Hugging Face forums.

How Do the Top Options Compare?

Prices reflect launch pre-orders. The base kit requires basic assembly. The Pro kit adds an NVIDIA Jetson Orin NX, extra sensors, and a larger battery. The research bundle includes teleoperation gloves and a 1 TB SSD. Sim Only provides software and model weights but no physical robot.

1. LeRobot Open Humanoid Base Kit , Best for teaching labs and first-time builders

LeRobot Open Humanoid Base Kit ships as a flat-pack assembly project. The box includes a 6-DoF torso, two 6-DoF arms, a differential drive mobile base, an RGB-D camera, and two-finger grippers. The robot uses a Raspberry Pi 5 with 8 GB of RAM. A 7-inch touch display sits on the chest. The base kit weighs 9 kilograms and reaches 1.2 meters tall. The onboard model is a 3 billion parameter vision-language-action policy with an 8,192 token context window. It runs at 10 Hz on the Raspberry Pi. The action rate is enough for pick-and-place and tabletop manipulation.

The software stack includes Linux, ROS 2 Jazzy, PyTorch, and the LeRobot training library. All code is Apache 2.0. Hardware files are CERN OHL-P. You can modify the CAD files, print replacement parts, or sell your own derived design. The base kit supports USB-C power delivery and standard 18650 battery packs. Battery life is about 90 minutes under active use.

The main limitation is the 3B model. It is not reliable for complex two-arm coordination. The maximum payload is 1.5 kilograms per arm. The base kit also lacks force-torque sensors. The rubber wheels work on smooth floors but struggle on carpet. For users who need stronger local inference, the Pro kit is the better option. You can follow the broader model release cadence on AI updates today.

Key strengths:

• ✅ Low $2,500 entry price makes humanoid robotics accessible to schools.
• ✅ Fully open Apache 2.0 software and CERN OHL-P hardware allow commercial use.
• ✅ Local 3B model eliminates cloud inference costs and privacy concerns.
• ✅ Repairable design uses standard parts and publicly available CAD files.
• ✅ Assembly guide and community forums lower the barrier for first-time builders.
• ❌ 3B model struggles with complex bi-manual tasks.
• ❌ Battery life is about 90 minutes under active use.
• ❌ No force-torque sensors on the base kit.

Who it’s for: Choose the base kit if you teach robotics or want an affordable open humanoid to modify.

2. LeRobot Open Humanoid Pro Kit , Best for research groups and robotics startups

The Pro Kit costs $3,900 and targets research teams that need more onboard intelligence. It swaps the Raspberry Pi 5 for an NVIDIA Jetson Orin NX with 16 GB of memory. That allows the 7 billion parameter policy model to run at 15 Hz. The context window remains 8,192 tokens for both model sizes. The Pro Kit adds force-torque sensors on both wrists, an extra rear camera, and a larger 6,000 mAh battery. The arms use higher precision motors with 0.2 degree repeatability. The mobile base has a higher torque motor option for light carpet use.

The 7B model performs better on long horizon tasks. In Hugging Face tests, it reached 72 percent success on unseen pick-and-place environments. The base 3B model reached 58 percent on the same test. That 14 point gap matters for research involving language conditioned tasks. The Pro Kit also includes a teleoperation mode using a smartphone or VR controller. You can record demonstrations and fine-tune the policy locally.

Software remains Apache 2.0. The Jetson module runs NVIDIA JetPack 6.1. You can deploy custom models from Hugging Face or train with the LeRobot library. The kit is still not a full industrial humanoid. It cannot lift more than 2.2 kilograms per arm. Runtime drops to 60 minutes when the 7B model runs continuously. But the price is far below closed platforms. If you are watching the shift from cloud APIs to local models, free AI pricing changes covers why local deployment is becoming more common.

Key strengths:

• ✅ More capable 7B model improves success rates on manipulation tasks.
• ✅ Jetson Orin NX enables 15 Hz inference for smoother control.
• ✅ Force-torque sensors provide feedback for contact-rich research.
• ✅ Larger battery and higher precision motors improve real world testing.
• ✅ Teleoperation tools help collect demonstration data quickly.
• ❌ Battery runtime drops to 60 minutes with continuous 7B inference.
• ❌ Still limited to 2.2 kilograms payload per arm.
• ❌ $3,900 price is higher than the base kit and less accessible for hobbyists.

Who it’s for: Choose the Pro kit if you are a research group that needs the 7B model and sensor feedback.

3. LeRobot Open Humanoid Research Bundle , Best for embodied AI labs that need full compute and teleoperation

The Research Bundle costs $5,500. It includes everything in the Pro Kit plus an NVIDIA Jetson AGX Orin 64 GB module, a 1 TB NVMe SSD, two extra depth cameras, and a pair of teleoperation gloves. The bundle also adds a mounting rail and a desktop charging dock. The extra compute lets labs run larger vision-language models or fine-tune the 7B policy on robot data. The context window for the included policy models is 8,192 tokens. Custom models can use longer contexts if the lab has enough vRAM.

This configuration is designed for data collection. The teleoperation gloves record joint angles, finger positions, and wrist force. Teams can build demonstration datasets and train on the local GPU. The LeRobot training stack supports LoRA and full fine-tuning. The extra cameras provide 360 degree perception for navigation and manipulation tasks. The bundle weighs 14 kilograms and includes a metal frame instead of the standard polymer parts.

The bundle is not a turnkey humanoid. It requires assembly, calibration, and software setup. Support is community based. The hardware warranty covers defects for 12 months. The main advantage is local compute. You do not need cloud GPU instances or API credits. That matters for labs with tight budgets. Agentic AI billing crisis explains why cloud agent costs can spiral. The Research Bundle is the closest Hugging Face gets to an offline embodied AI workstation.

Key strengths:

• ✅ Jetson AGX Orin 64 GB handles larger models and fine-tuning.
• ✅ Teleoperation gloves accelerate demonstration data collection.
• ✅ 1 TB SSD stores large robot datasets without external drives.
• ✅ Extra depth cameras improve navigation and manipulation perception.
• ✅ All local compute eliminates recurring cloud GPU or API fees.
• ❌ Assembly and calibration are more complex than the base kit.
• ❌ Higher $5,500 price limits use to funded labs.
• ❌ Community based support may be slow for hardware issues.

Who it’s for: Choose the Research Bundle if you need local fine-tuning, teleoperation, and larger datasets.

4. LeRobot Open Humanoid Sim Only , Best for developers who want the stack before buying hardware

The Sim Only option is free. It includes the complete software stack, 3B and 7B model weights, and simulation assets for MuJoCo and Isaac Sim. Developers can train policies in simulation and evaluate them before buying a physical kit. The model weights use the same Apache 2.0 license as the physical robot software. The context window for the policies is 8,192 tokens. The package includes 20 sample tasks, from pick-and-place to drawer opening.

The simulation is not a perfect replacement for reality. Sim-to-real transfer remains a risk. Policies that work in MuJoCo may fail on real hardware due to friction, lighting, and motor backlash. The LeRobot team provides a gap report with common failure cases. The Sim Only option also includes the ROS 2 packages so developers can integrate with existing robot stacks.

This free tier is important. It lets students and independent developers learn the software without a $2,500 hardware purchase. The models run on a desktop GPU with at least 8 GB of vRAM for the 7B version. The 3B version runs on CPU-only machines at slower rates. For a broader view of free model and tool access, see free AI pricing changes in June 2026. The Sim Only option is not a product in a box. It is an open door to the same code that runs on the physical robot.

Key strengths:

• ✅ Free access to the full software stack and model weights.
• ✅ No hardware purchase required to start developing.
• ✅ MuJoCo and Isaac Sim support covers popular research simulators.
• ✅ Apache 2.0 license permits commercial use of the simulation code.
• ✅ 20 sample tasks help new users learn quickly.
• ❌ Sim-to-real gap means policies may fail on physical hardware.
• ❌ 7B model needs at least 8 GB of GPU vRAM for reasonable speed.
• ❌ No physical robot data for calibration and real world testing.

Who it’s for: Choose Sim Only if you want to test the stack or train policies before committing to hardware.

Frequently Asked Questions

What Should You Remember?

• Price: The base kit costs $2,500, far below closed humanoid platforms.
• Open licenses: Software is Apache 2.0 and hardware is CERN OHL-P, so commercial use is allowed.
• Onboard models: Local 3B and 7B policy models use an 8,192 token context window.
• Repairability: All CAD files, PCB designs, and firmware are public for modifications.
• Trade-offs: Payload is 1.5 to 2.2 kilograms per arm and battery life is 60 to 90 minutes.
• Availability: Pre-orders started June 10, 2026 with shipping in July 2026.
• Simulation: The free Sim Only package lets developers train before buying hardware.

Free AI News is an independent editorial publication. Information about AI pricing, free-tier limits, and features changes frequently and may become outdated. Always verify current details through the vendor’s official pages. Affiliate links may earn a commission at no cost to you, and never affect our reporting.