Google DeepMind Genie 4

Track; not yet

Why

Genie 4 is, on DeepMind's own account, its first real-time world model to generate audio natively, and it takes mid-simulation text steering, with Gemini 3.8 Flash as a guiding agent in the loop. The capability claims, however, rest on the vendor's announcement and demos: no benchmarks, latency or cost are published, access is limited to hand-picked testers, and DeepMind concedes object-permanence drift and unsolved text and audio accuracy. The two-minute consistency figure comes from third-party press (TechCrunch reports it and Online Tech Tips attributes it to Android Authority) and is not in Google's post. There is nothing for a production team to integrate yet, but the direction matters for game prototyping, film previsualization and agent or robot training environments.

Next

Watch for the Project Genie (Google AI Ultra) and Gemini API / AI Studio rollout, published session-length, latency and cost figures, and independent hands-on or benchmark tests; re-evaluate at general availability.