Boston Dynamics ATLAS Exposed 2 New Football GAME CHANGERS ($13,000 HUMANOID)

Is bots and dynamics the new master of humanoid movement? Today on AI News, we're taking a closer look at two new skills that Atlas just learned, plus two new skills from AGI Bots X2 for you to decide which robot is learning faster. Starting with the LingXi X2 from AGI Bot because they just recreated Atlas's fridge task. But before I show you that, let's first see their autonomous AGI Bot Generative Intelligence Locomotion Engine, or AGI for short, which they say gives this X2 a brand new level of visual awareness. And here you can see it's dodging these balls completely autonomously. But something that's even more impressive here is this recreation of what Atlas just did last week. It picked up a refrigerator with over 100 pounds of loose items inside. Now the X2 is autonomously picking up a refrigerator of its own, although we don't know in this specific demonstration how much weight is inside, if any at all, or is it loose? Also we can see here, is this cheating? Tell us in the comments down below. There are two handles that are latched into this robot's arms. But nevertheless, despite being just 4.3 feet tall or 131 centimeters, it was able to successfully carry this refrigerator to the target space. But what else would you like to see from this X2 to prove that it's really learning? Meanwhile, Boston Dynamics is pushing the limit even further, teaching its Atlas robot two brand new skills. First here is a kick. And this is after the robot already learned just a few days ago how to do footwork, feint, drill, and even celebrate. So now this robot is learning how to actually drive this ball into the goal with your standard kick. You can see it seems to have made this goal. So then Boston Dynamics ups the ante here with a second skill, which is the Rabona, a special kicking move where the leg goes behind. And it did this just by watching, first with a normal shot, figuring out exactly how to orient its body, first without kicking the ball with its primary leg, and then skipping over with its back leg to hit that fake shot. And we can see here with its grippers, it's even holding this ball. Would be good to see some more dexterous manipulation, but let's take a look again at this Rabona kick. And this is another video that they just released where they're showing this up close one more time here as it was messing up with its practices. And finally, it nails this Rabona kick. So comment down below, what should be the next skill that this robot should learn? Maybe not just for soccer, but what about a dexterous manipulation task or even a soccer goalie move where it could use its hands as well as feet to block incoming projectiles like AGI Bots X2. And here we just have a demonstration of both of these robots side-by-side, autonomously picking up and moving these refrigerators. Tell us down below, which one would you trust more in your home, assuming that you weren't there to supervise it. And while you watch this, just for reference, the X2 is projected to cost between 13 to $55,000 per unit. Whereas Boston Dynamics is still currently looking at somewhere around 130 to $140,000 per robot with the cost potentially dropping down nearer to $30,000 as they scale with production. But in a Gentik AI, Google DeepMind just introduced Gemini 3.5 Flash, shifting its AI away from a single prompt assistance to high-velocity multi-agent factories. In fact, when looking at output tokens per second, 3.5 runs four times faster than other frontier models at less than half the cost in order to eliminate the speed versus intelligence bottleneck. Furthermore, by leveraging Google anti-gravity infrastructure, developers can now deploy orchestrated networks of collaborative sub-agents in parallel, allowing them to use many agents at once. And because of this, a coordinator agent can now break down open-ended goals and assign different tasks to specialized sub-agents simultaneously, allowing the swarm to execute complex, multi-week-long workflows in just minutes. And it can all be done while maintaining frontier performance with some of the many examples, including anti-gravity and 3.5 Flash, executing an automatic renaming and categorization of unstructured assets based on dynamic criteria. And you can see here that this would otherwise have taken quite a long time. Whereas in another example, they use anti-gravity and 3.5 Flash to synthesize AlphaZero's paper and then code a fully playable game in just six hours. Or else they also use 3.5 Flash and anti-gravity to transform an entire legacy code base into Next.js. But one of the coolest things is that they had these AI agents create entire cities as a swarm. And you can see here what they came up with. This is an autonomous city builder that came to life all on its own. Meanwhile, 3.5 Flash can also create interactive animations for research papers on AI Studio, as well as take plain text descriptions and turn them into interactive hardware on AI Studio. But while Google focuses on high-speed efficiency with Gemini 3.5 Flash, Anthropic is taking a different approach to multi-agent scaling by instead upgrading its raw frontier intelligence. So Anthropically just shipped Claude Opus 4.8, which they describe as a modest but tangible improvement with their flagship model successfully topping OpenAI's GPT 5.5, as well as Google's Gemini 3.1 Pro across most industry benchmarks. But the true highlight of this release is the introduction of dynamic workflows, where similar to Google's anti-gravity framework, this feature now allows Claude Opus 4.8 to autonomously schedule tasks and deploy hundreds of specialized parallel sub-agents simultaneously. With this architecture tightly integrating into platforms like Claude Code, where they enable sub-agents to write code concurrently inside isolated cloud environments, execute bug fixes, as well as manage massive code-based migrations. And on top of this, Anthropic just introduced a new effort control setting, which lets users choose exactly how much computing power the model dedicates to each given query, with pricing staying the same as the previous version at $5 per million input tokens and $25 per million output tokens, though a new fast mode does reduce operational costs. Anyways, check out this video of this robot juggling and comment down below whether or not you could do better. And don't forget to hype, like, and subscribe, as well as click on this video here to see more of the latest in AI and robotics news.