Can humanoid robots cook?

The short answer: not yet

No humanoid robot in 2026 can cook autonomously at consumer-deployment scale. Demonstrations of food-handling exist across the cohort; research-level cooking robots exist at the purpose-built tier (Moley Robotics's installed kitchen platform); a general-purpose humanoid that walks into a customer's kitchen and prepares a meal is not a 2026 product.

Per DEPLOY's verified-vs-claimed framework on capability claims, cooking is a worked example of the gap between marketing demonstrations and consumer-deployment verification. Task complexity (object recognition + heat and tool handling + multi-step coordination + safety + cleanup) is substantial; the verification surface for "can it cook" has not been cleared by any cohort manufacturer at consumer-deployment scale.

What the cohort actually demonstrates

Applying the framework across humanoid manufacturers:

• 1X NEO: consumer-deployed with explicit teleop disclosure. NEO performs simple manipulation tasks in consumer homes (folding clothes, organizing); cooking is not advertised as scope. The verified-deployment scope for NEO is light household task support with operator-assisted complex tasks, not autonomous meal preparation.
• Figure AI (Figure 02 + Figure 03): enterprise-deployment focus (BMW Spartanburg + Catalyst Brands Reno). Cooking is not in Figure's product scope; commercial customer base is automotive manufacturing and logistics, not consumer kitchens.
• Apptronik Apollo: enterprise pilots at Mercedes, GXO, and Jabil. Cooking is not in product scope; deployment envelopes are manufacturing + logistics.
• Tesla Optimus: video demonstrations exist showing food-handling (serving drinks at We Robot 2024; some object placement). Consumer-deployment cooking is not verified; Tesla has not opened consumer orders for Optimus.
• Boston Dynamics Atlas: research and elite-R&D context; cooking is not in scope for the commercial deployment path Boston Dynamics is pursuing.
• Unitree G1 + R1: research-tools positioning. Cooking is not in product scope; platforms ship to research and developer customers.

Cohort-wide read: zero humanoid manufacturers are selling cooking capability as a current product. Cooking demonstrations that exist in marketing content are demonstration-grade, not consumer-deployment-grade.

Why cooking is hard for humanoids

Cooking is exactly the verified-vs-claimed exemplar at task layer. Task decomposition reveals why:

• Object recognition under variable conditions: ingredients vary by brand, freshness, size, and visual presentation. A robot policy that recognizes a tomato in lab conditions may not recognize the same tomato in a customer's kitchen with different lighting, variable container, different background.
• Heat handling and safety: stovetops, ovens, hot pots, and steam are safety-critical interactions where misjudgment produces physical harm to the robot, the user, the kitchen, or all three. Robot insurance and liability frameworks for in-home heat operations are not established.
• Multi-step coordination: a recipe involves sequencing multiple sub-tasks (chopping while monitoring heat; transferring while timing; cleanup while preparation continues). General-purpose humanoid policy stacks have not demonstrated this kind of long-horizon task coordination at deployment reliability.
• Cleanup: kitchens generate waste, dirty utensils, dropped food, and spilled liquids. Cooking capability without cleanup capability produces a robot that creates more work than it completes.
• Customer-specific kitchens: every home kitchen has different layouts, appliances, cabinet positions, and utensil organization. A robot that cooks in a controlled demo kitchen may not cook in a customer's actual kitchen without per-installation calibration.

Purpose-built kitchens versus humanoid generalization

The market does have one verified consumer cooking robot path: purpose-built robotic kitchens that install permanently. Moley Robotics offers a robotic-arm-based installed kitchen at approximately $300,000 (full kitchen installation, not a humanoid that walks in). Samsung and several appliance makers have shown kitchen-robotics concepts. These products are not humanoids; they are installed kitchen appliances with robotic capabilities.

The trade-off:

• Purpose-built: works in the kitchen it's installed in; doesn't transfer to other homes; high upfront cost; limited to cooking and adjacent tasks.
• Humanoid: would in principle work across many homes if cooking capability worked; cooking capability does not yet work at consumer-deployment scale; the trade-off has not yet been resolved.

What's plausible by what timeline

Per DEPLOY's framework on claimed-future capability: humanoid cooking at consumer-deployment scale is multiple verified-capability iterations away. The closest verified consumer humanoid capability (1X NEO at light household tasks with teleop disclosure) is several capability layers below autonomous cooking. Whether the gap closes in 5 years or 15 is operator territory; the framework does not claim either timeline.

For consumer purchase evaluation: nobody currently sells a humanoid robot that cooks. Anyone marketing a humanoid that cooks autonomously is selling research and demonstration, not consumer-deployment capability.

Where to go for context

For the full capability-disclosure framework across humanoid manufacturers, see what can humanoid robots actually do today. For the specific consumer-deployed humanoid capability that does exist (light household tasks with teleop), see is 1X NEO autonomous or controlled by humans. For the Tesla Optimus task-level capability evaluation, see Tesla Optimus capabilities.

For consumer-evaluation context across the cohort: 1X NEO, Figure 03, Tesla Optimus pricing pages. For methodology canonical references applicable to capability-task framing: verified-vs-claimed at within-entity granularity (demonstration-vs-deployment per-claim depth) + the 4-way autonomy-boundary taxonomy.