New GENE 01 Bionic Humanoid Robot Teases AMD AI Game Changer (KIMI K3 VS FLASH 3.6)

Generative Bionics just previewed its upcoming humanoid with several new tech features, but can it beat Tesla and Boston Dynamics? Plus, today on AI News, we'll compare Flash 3.6 with Kimi K3 for you to decide whether AI's future is open or closed source. But first, did you know you can earn hundreds of dollars per hour discussing your expertise with AI? Let Mercore or Micro One start compensating you today by clicking the link below so that AI News can vouch for you and have your first conversation. But now, back to the news, as Generative Bionics just previewed its Gene01 AI robot using touch as a source of intelligence. It does this by combining tactile force and vision feedback at high frequency to feel contact, understand intent, and then react safely and appropriately, which Generative Bionics says turns interaction into Gene01's core capability instead of just another feature. In fact, it has multimodal skin covered by an anti-pinch material for safer human collaboration, with this distributed network of tactile skin sensors detecting touch, proximity, force, and temperature, so that Gene01 doesn't just understand, but can even anticipate human presence, as well as objects in its environment. Plus, on top of touch sense and proximity, it features 24 degrees of freedom in its hand and wrist, which is complemented by Gene01's surround vision, but details are sparse as to whether these are LiDAR or fisheye cameras, and how many there are throughout its body. Also tell us in the comments whether you think this 3D mesh in its hands is made of thermoplastic elastomer or silicon or something else to achieve what looks like an impressive level of flex. But most incredibly, this is an open access platform as Gene01 comes with an open source digital twin description, making it the first humanoid robot to ever be published and maintained across ROS, PyPi, and CondaForge to be used in physical AI, allowing developers to simulate, build on, and test before Gene01 is ever deployed, and the company will be competing directly with Tesla and Boston Dynamics as they're aiming for factory, heavy industry, and healthcare applications to start. In fact, Generative Bionics is from Italy, and they claim they've executed the fastest ever transition from concept to a fully functional humanoid robot in just six months, and now Gene01 is integrating body mechanics and motion intelligence into one system to achieve physics native AI. Furthermore, because Generative Bionics is focusing on building a scalable platform for manufacturing, its Gene01 can be customized across its AI, exterior design, and even its end effectors to fit specific industrial applications. But how much do you think this should cost, and will they be able to meet their Q4 2026 launch schedule? If so, they could scale out their manufacturing of Gene01 to create thousands or tens of thousands of units at a selling price of between $20,000 to $30,000. But to get there, they'll need ultra-dextrous hands, which may become swappable between robots, allowing solutions like the M1 Hand to drop the overall research and development costs of ultra-dextrous humanoids over time, as Mimic officially unveiled its brand new Mimic M1 Hand alongside the Mimic wearable U1 teleoperation system to solve general purpose dextrous manipulation. Instead of adopting a locked humanoid design, Mimic takes a platform agnostic approach, allowing its M1 Hand to deploy on a single or dual arm stationary table, or attached to autonomous mobile platforms to handle tedious operations like manufacturing and logistics. And what's more, is that Mimic purposefully duplicates the human hand morphology because it unlocks the ultimate data stream, internet-scale video of human interactions. And this data-driven strategy allows their end-to-end learning framework to automate unstructured industrial work, handling changing positions, irregular objects, soft bodies, and cables that were previously impossible to automate. And to power this next generation of dexterity, Mimic breaks away from the traditional vision language action model, and instead uses a pioneering vision action model that grounds robotic policies directly into pre-trained video world models. And they've even partnered with Black Forest Labs to launch FluxMimic, scaling these native video action models so that robots can intuitively learn complex manual assembly, packaging, and sorting tasks with absolute minimal fine-tuning. But while physical AI companies are making different bets on hardware, two massive AI labs also just made two very different bets on how intelligence will be delivered, as Moonshot AI, the Beijing lab behind Kimi Assistant, just released its Kimi K3, a massive open-source model that anyone can download, while simultaneously Google's DeepMind just shipped Gemini 3.6 Flash as a closed-source, API-only upgrade that's built for speed and cost, with the contrast highlighting two possible futures for AI, being open-source or closed-source. So first, Kimi K3 is, by Moonshot's own description, the world's first open model in the 3 trillion parameter class at 2.8 trillion parameters, and it runs on a new architecture called Kimi Delta Attention paired with Attention Residuals, activating just 16 of 896 experts per token through a sparse mixture of experts' design. This efficiency, Moonshot says, lets K3 turn compute into capability roughly 2.5 times better than its predecessor. Plus, the model natively handles images and video alongside a 1 million token context window, and Moonshot's demos show it writing a GPU compiler from scratch, designing a physical chip, and editing video, with Moonshot saying K3 trails the very top proprietary systems overall, but is competitive with them benchmark by benchmark, with full waits being due by July 27th. Then there's Google's DeepMind Gemini 3.6 Flash, which is taking the opposite approach, no downloadable waits, just an API upgrade for Google's developers. It replaces Gemini 3.5 Flash as the company's mid-tier workhorse, needing 17% fewer output tokens for comparable work, with the output pricing cutting from $9 to just $7.5 per million tokens, and gains were substantial in both coding and computer use. But when compared directly, Kimi K3 leads on shared benchmarks, including DeepSWE and TerminalBench 2.1, but Gemini 3.6 Flash is cheaper, costing just $1.5 per million input tokens and $7.5 per million output tokens versus K3's $3 per million input tokens and $15 per million output tokens. Plus, K3 supports 1 million token outputs versus Flash's 64,000 token cap, but K3's waits aren't downloadable until July 27th and require the expensive hardware to self-host, while Flash is live right now and available via API. So tell us what you prefer. Anyways, don't forget to share your skills in the comments below and we'll vouch for you to be hired on Mercore as referred by AI News. So click the link, like and subscribe, and check out this video here for more of the AI and robotics news and thanks for watching!