China’s DeepSeek has been making waves in the AI world with its lightweight, high-performance models that run efficiently on low-cost hardware like smartphones. Now, Google is hitting back. On Wednesday, the tech giant unveiled Gemma 3, its latest open AI model, signaling that it’s still very much in the race to dominate the next-gen AI space.
With four model sizes ranging from 1 billion to 27 billion parameters, Google claims Gemma 3 outperforms DeepSeek-V3, Meta’s Llama-405B, and OpenAI’s o3-mini, making it the best open model that can run on a single AI accelerator chip. But that’s not all—Google is also making a fresh push into robotics AI, bringing Gemini 2.0 into the world of autonomous machines and advanced spatial reasoning.
So, is Google’s latest AI play enough to challenge DeepSeek’s dominance? Here’s what you need to know.
Gemma 3: Google’s answer to DeepSeek’s lightweight AI
Gemma 3 comes in four sizes, ranging from an ultra-compact 1 billion parameters to a still-efficient 27 billion parameters, making it ideal for mobile devices, single-chip AI accelerators, and lightweight computing environments.
Beyond efficiency, Google claims Gemma 3 is the most advanced open model yet, boasting a 128,000-token context window—finally catching up with DeepSeek and Meta’s Llama in memory capacity. If these claims hold up, Gemma 3 could be Google’s best shot at reclaiming dominance in open AI models.
Google’s AI safety push
To address growing concerns around open-source AI safety, Google has also introduced ShieldGemma 2, a content moderation tool designed to detect violent, explicit, or harmful images in AI-generated content.
Google is emphasizing that Gemma 3 has been rigorously tested, particularly in STEM-related areas where AI could potentially generate harmful instructions. With DeepSeek’s rapid expansion, Google seems keen to differentiate itself by focusing on responsible AI deployment.
Google’s big robotics play
AI isn’t the only battlefield—Google is also making a fresh push into robotics. With DeepMind’s Gemini 2.0, Google has developed two specialized robotics models:
-
Gemini Robotics: Capable of turning natural-language instructions into physical actions, adapting to environmental changes. Google claims it can fold origami and pack Ziploc bags, a significant step in robotic dexterity.
-
Gemini Robotics-ER: Focuses on spatial reasoning, allowing robots to analyze and grasp objects correctly—like picking up a coffee mug at the right angle without spilling it.
This marks Google’s first big leap back into robotics after shelving its Everyday Robots project, suggesting it still sees a future in AI-powered automation.
Google vs. DeepSeek: Who’s winning?
With DeepSeek pushing efficient, high-performance AI models, Google’s Gemma 3 is its counterstrike—offering a powerful open model that can run on minimal hardware. At the same time, Google’s robotics push positions it against companies like Boston Dynamics and Tesla Optimus, showing that it wants to lead in both software and hardware AI applications.
But will Gemma 3 and Gemini Robotics be enough to outshine DeepSeek and other rising AI players? Or is Google still playing catch-up in the open AI race?
Google isn’t backing down
With Gemma 3, a renewed robotics strategy, and a stronger focus on AI safety, Google is making it clear—it’s not letting DeepSeek or other AI challengers steal the spotlight. Whether Gemma 3 can truly outperform DeepSeek-V3 remains to be seen, but one thing is certain—Google isn’t done fighting for AI supremacy just yet.