Figure AIs 03 vs Apollo 2: AI Robot Gamechangers (OMNI SEEDANCE 2.5)

You can see it manipulating these packages here. That’s eight cameras in total, using its tactile sensors, sensing as little as three grams in each of its proprietary fingertips, and in terms of its AI, what you’re seeing here is Helix 02, which is a three-system approach autonomously charging this robot with wireless inductive coils built in to its feet, and when it comes to real-world execution, here are six different tasks. Five of these are Figure 3. They actually were able to sort over 40 hours straight, uninterrupted, with each robot serving for five hours at a time, as well as perform kitchen cleaning tasks and general house care tasks across different items, even in tandem with two robots working together with each other over the air to fold this blanket, which hasn’t been replicated yet by other robot companies, and when it comes to the expected price and release date, so far they’ve built more Figure 3s than they have employees at Figure, over 700 in fact, and each of these robots, they’re aiming for a price of about $20,000, with some of the latest reports being around $24,000, and they plan to have over 100,000 robots in the next four years, with a deployment strategy of late 2026 for a wider set of non-commercial partners, but let’s see how all of this compares to the newest Apollo 2 from Apatronic.

This robot can handle a payload capacity of up to 55 pounds or 25 kilograms, and it actually walks at 2.1 miles per hour, and it’s already working at Mercedes-Benz, as well as with Google DeepMind, with LIDAR, and onboard NVIDIA hardware for 275 tera-operations of AI processing, and this robot actually takes over-the-air updates, with it already working at Robot Park, which is Apatronic’s flagship 90,000-square-foot data collection and training facility, and so far, in terms of their expected price, they’re looking at a long-term target of about $50,000, with continued commercial-side pilots through the end of 2026, and possibly a wider rollout to regular customers in 2027.

Plus, there’s actually a third robot that just came out from AGIbot named the A3, and here is what the director of physical AI had to say about it. And in terms of specs, so far we know that this weighs just 55 kilograms, and it can function for 10 hours long using its 10-second battery swap system, with the ability to work in swarms that are synchronized between 100 separate robots. Then, in terms of hardware, it features shoulder tactile sensing, but the rest of the info around this robot is still unavailable as AGIbot is continuing to release this slowly, so comment down below if you have any more information and also what you think this A3 should cost, or if it would be better to rent or just wait for the A4.

At the same time, in the generative AI world, Google OmniFlash and SeedDance 2.5 are now head-to-head, with OmniFlash now in public preview for all users in Google’s AI studio. In fact, OmniFlash provides even faster, cheaper video generation and conversational editing. Plus, its multimodal support accepts text, image, audio, and video, so you can feed the model text prompts combined with reference photos, audio clips, and existing videos simultaneously. And Omni also handles multiple output formats, native audio generation, and complex video synthesis from just one prompt. Currently, it’s priced at 10 cents per second, with a multi-turn conversational video editing style that allows for adjustments of any detail, track text, or anchor graphics with plain language, and generated videos are up to 10 seconds long at 720p, 1080p, or 4k resolutions.

But which is superior, because ByteDance also just officially introduced its own SeedDance 2.5 on its Dreamina platform, which is a major update to the inclusion of region-to-video or R2V reference tracking, which allows creators to use green screens or even white models to map character spatial locations, as well as structured motion paths, rather than relying solely on text prompts. On top of this, the model significantly extends clip durations, with the standard mode offering 30-second video clips, and the beta long video mode offering 180-second clips, all while preserving character identification, lighting, and scene consistency.

And to streamline workflows, SeedDance 2.5 also supports up to 50 multimodal inputs including text, scripts, photos, video, and music within a single generation process. But tell us down below whether you prefer SeedDance 2.5 or Gemini, OmniFlash, and why. And like and subscribe, as well as check out this video, and thanks for watching. [tr:trw].