Google Unveils Two AI Models: Argon Scores 87% in Physical Reasoning, Genie 4 Generates Worlds in Real Time

Google just released two models, both connected to the "physical world."

Gemini 4 Argon: 87% in Physical Reasoning

Gemini 4 Argon is a multimodal reasoning model with enhanced physical understanding, code generation, and visual semantic reasoning. On the embodied reasoning benchmark ERQA, it scores 87% through pure reasoning, leading other frontier models. Benchmarks like ERQA test AI's understanding of objects, space, and physical laws—they're not easy to game.

Argon can also direct Boston Dynamics' Spot quadruped robot to complete tasks such as retrieving objects and throwing. The simulation data it generates improves real robot success rates from 56% to 75%, and cuts single-task time from 35 seconds down to 24 seconds. The scarcest resource in robot training is data—real-world collection is expensive and slow, making synthetic data a critical path.

The model is available via the google-genai SDK. Pricing is set at $2 per million input tokens, with output pricing not fully listed.

Genie 4: Real-Time Generation of Interactive Worlds

The other release is Google DeepMind's world model, Genie 4. It generates visuals, ambient audio, and interactive dialogue in real time, and lets users modify scenes at any point using natural language. Gemini 3.8 Flash serves as the guiding agent behind it, breaking down complex requests into multi-step tasks that are then handed to the generative model for execution.

For now, Genie 4 is only available as a research preview to trusted testers. It will later open up to Project Genie Ultra subscribers, as well as partners using the Gemini API and AI Studio. Generated content carries a SynthID watermark to identify AI-produced output.

Looking at the two models together, Google's direction is pretty clear: just chatting isn't enough—AI needs to interact with the physical world. Argon fills the data gap for robots, while Genie 4 lowers the barrier to creating virtual worlds. These are still early days, but the direction is far more interesting than another round of conversation.