Figure 03 vs Tesla Optimus 2.5: AIs Humanoid Robot End Game (CHATGPT APPS)

Next, the hands have been covered with a grippier surface for better performance in pick and place, as well as the addition of a camera and what look to be potentially sensors on the fingertips. But tell me what you think these are. Next, it looks like you can take apart its joints with these tabs. It doesn’t look like you could get your fingers caught in these edges, for instance. Maybe there you could. Next, you can see that the garment is actually tucked up under the helmet. I’ll tell you something about this helmet in just one moment.

Another look at this material. It would be interesting to know what it’s made out of, and is it fire-retarded, for instance. Here’s the waist joint. This is the back with presumably heat ventilations and a power button. You can unzip the robot to reach its components. Now, look at this, a battery display on its head. And on the back of the waist, we can see more heat dissipation vents. The exploded view of its componentry. See some LED lights there. What resolution do you think those fingertips are? Another view of the camera inside the palm of the hand, which prevents occlusions from the fingers.

And here, the foot, you can see it has articulating toes much like the Tesla, but it also has inductive charging on the heel, which means that the robot can charge by simply standing. And it comes with apparently different types of apparel that you can choose for it. A back-facing camera. And notice this power display. It tells you the battery charge of the robot. And this is the latest footage of the robot working for 10 hours at a time in the BMW plant. This is the figure two. This is already working in the real world, completely autonomously picking and placing these parts.

Tell me what you think comparatively to Tesla’s Optimus, which we’ll get into in a second, which is further ahead from a commercial standpoint. Brett Adkopf said that production will likely start in 2026 or a wider release for consumers, saying that the robot’s target price is between $20,000 to $30,000 at scale. But tell me how long you think that it’ll take to get there. But how does Tesla Optimus compare to figure three? Let’s see. Elon Musk says that this clip was fully AI-powered, meaning that there’s no teleoperation. This was completely autonomously controlled by reinforcement learning.

So the robot here is probably following some type of a dance that it learned from human instruction, but it’s carrying it out in real time, seemingly in reaction to this mocap actor’s movements. Look at how it readjusts its balance here. One thing to note here is you look at the feet, and you can see its articulating toes as it moves forward. It really helps with the balance, and it pushes it back, again, rebalancing itself using those articulated toes. And then there’s another clip of the Tesla Optimus robot carrying out similar kung fu moves here on the red carpet.

What’s very impressive here are its toes specifically. Look at the way that its feet bend. This really helps with the balance in real time. But tell me in the comments below whether or not you think that this is completely AI controlled, or if it’s just teleoperation for the red carpet. But AI might be coming for more than just your robots, as it might control your mouse and your keyboard soon too, as Google just launched Gemini 2.5 computer use model on top of Gemini 2.5 Pro’s visual understanding and reasoning. And the goal here is to create AI agents that can interact with user interfaces just like humans do, like by clicking, typing, scrolling, and more.

And currently, Gemini 2.5 computer use is mostly optimized for controlling web browsers, but it’s also showing strong early results on mobile user interface tasks, though it’s not yet tailored for operating system-level desktop control. And the model’s design allows it to natively handle complex digital tasks, for example, it can fill out or submit forms, manipulate drop-downs and filters, and even navigate behind login screens. Developers can interact with these features using the new computer use tool in Gemini API, and they can provide the model with the user requests, a screenshot of the environment, and even a history of recent actions.

Then developers can limit or expand the set of actions the model can take, allowing for finer grain control. And when deployed, the model analyzes the provided contacts and generates function calls, representing the specific user interface actions that they want to carry out. But certain sensitive actions, like making purchases, will trigger requests for end-user confirmation. And after each action, the environment is updated and the loop continues until the task or workflow is complete and error occurs, or safety protocols halt the process altogether. And performance-wise, Gemini 2.5 computer use is outpacing its competitors on benchmarks, including tests run by browser-based and Google’s own teams.

But there may be another competitor in the agentic race, as OpenAI just released its newest apps and chat GPT feature to create, learn, and automate tasks. This is all inside of chat GPT using interactive apps that are built directly into the chat experience. And now users inside the EU on free, go, plus, and pro plans will all have access to curated sets of apps to integrate within their conversations. And OpenAI made sure to highlight how these apps can interact. Users can invect them directly by name, or they can discover them when chat GPT suggests relevant apps based on context of the chat.

And these apps then respond in plain language, with many including interactive elements like maps, playlists, and presentation tools, which are all accessible within the chat window. On top of that, chat GPT is also launching its apps SDK across the world. And the SDK is built on top of model context protocol, which is the open standard for connecting generative models with external tools and data. And this allows developers to design custom logic and interfaces, plus they can connect directly to their own back ends for authentication and premium features. And with more than 800 million chat GPT users, OpenAI says that user safety and privacy is still a top priority with every app adhering to strict usage guidelines and privacy policies, collecting only essential data and being transparent about permissions.

And the first time a user connects to any app, clear prompts will explain what information may be shared, plus enhanced privacy controls are on the way in the near future to let users have even greater control over their data options. But share down below whether or not you’d use N8N, which is open source, or OpenAI, which is more of a closed source alternative. Tell us down below, make sure to like and subscribe, and check out this video here for more on the latest in AI news. [tr:trw].