Google DeepMind introduced Genie 4, a real-time interactive world model with native multimodal audio (ambient, music, interactive dialogue), steered via natural language and built on Gemini Omni + other Gemini models for coherence. Research preview to trusted testers; coming soon to Google AI Ultra / Project Genie waitlist and early-access Gemini API + AI Studio partners. SynthID watermarking on generated video/audio. Company acknowledges limitations (object drift, audio accuracy, in-world text, long-run consistency). Soft title: first-party research-preview world-model ship — not GA Ultra/API yet; vendor-reported capabilities. Stamp: DeepMind + Google blogs Oct 27; @GoogleDeepMind ≈ 00:10 HKT Wed 28.
Same-morning frontier world-model surface for agent/robot training sims and creative tooling — Models lane lead.
More ▾ Less ▴
DeepMind and Google first-party blogs dated Oct 27, 2026 introduced Genie 4 as a real-time interactive world model with native multimodal audio, steered by natural language and built on Gemini Omni plus other Gemini models. @GoogleDeepMind amplified ≈ 2026-10-27 16:10 UTC (≈ 00:10 HKT Wed 28). Availability is research preview to trusted testers, with Google AI Ultra / Project Genie waitlist and early-access Gemini API + AI Studio partners framed as coming soon. SynthID watermarking applies to generated video/audio. The company lists limitations including object drift, audio accuracy, in-world text, and long-run consistency. Soft title: research-preview world-model ship — not GA Ultra/API yet.
- Ultra / API GA timing and pricing
- Robotics/sim customer deployments
- Independent coherence and long-horizon evals
- SynthID detection uptake
Vendor-confirmed research preview; capabilities and benches not independently reproduced here.