If you thought the Pixel 10 launch told you everything about Tensor G5, think again. Google’s latest Made by Google podcast dives into the chip’s hidden tricks that are ushering in the era of the “AI phone”, and it turns out there’s way more going on under the hood than glossy keynote slides suggested.
According to Jesse Seed, Group Product Manager for Google’s Silicon Team, the Pixel 10 is quietly running a nearly 4-billion effective parameter Gemini Nano model entirely on-device. That’s a staggering leap for mobile AI, made possible by co-design work between DeepMind and Google’s Tensor team that started well over a year ago. The pair went further, introducing a new “Matformer” architecture, which is essentially a single LLM that can switch between a smaller sub-model for speed and the full version for quality, depending on the task at hand.
The camera story is just as wild. ProRes Zoom isn’t simply “better algorithms.” It’s powered by a diffusion model with nearly a billion parameters running directly on the Pixel 10. Jesse admits early builds took over 100 seconds per shot, but thanks to Tensor G5’s 60% TPU uplift and system-wide optimizations, those same shots now render in just seconds. That’s how the Pixel 10 Pro can suddenly pull off 100x zoom shots with real detail.
Another shocker? Voice Translate doesn’t need any prior enrollment to clone your voice. In real-time, the Pixel can translate both sides of a conversation, and your voice is preserved without uploading samples or training. Google also confirmed Recorder app summaries are now 2.6x faster and twice as energy efficient versus last year, a direct payoff from Tensor G5’s upgraded CPU, memory, and storage interfaces.
Google keeps saying it designed Tensor not for benchmarks but for real-world AI features, and these podcast revelations drive that point home. From massive on-device LLMs to diffusion photography and instant voice cloning, Tensor G5 feels less like a spec bump and more like a quiet line in the sand. Pixel is no longer just a smartphone. It’s Google’s vision of what an “AI phone” really looks like.
Hillary Keverenge
2700 Posts
Tech has been my playground for over a decade. While the Android journey began early, it truly took flight with the revolutionary Lollipop update. Since then, it's been a parade of Android devices (with a sprinkle of iOS), culminating in a mostly happy marriage with Google's smart home ecosystem. Expect insightful articles and explorations of the ever-evolving world of Android and Google products coupled with occasional rants on the Nest smart home ecosystem.
Next article View Article
