Genie 4

Spec

Attribute	Value
Developer	Google DeepMind
Released	2026-10-27 (trusted-tester research preview)
Announced	2026-10-27
Context window	unknown
Pricing	unknown
License	unknown
Availability	Trusted testers; Project Genie, Gemini API and Google AI Studio access forthcoming
The announcement establishes the restricted preview, not general availability. (source)

Release Date

Google announced the preview on 2026-10-27. Broader access is planned first for Google AI Ultra subscribers in Project Genie and early access partners of the Gemini API and Google AI Studio; the post gives no date for that expansion. (source)

Benchmarks

The launch describes interactive demonstrations and tester feedback, without a numerical comparative benchmark in the captured prose. Google acknowledges drift in visual detail and object permanence, especially in longer sessions or after complex edits; physics, audio accuracy, legible text and spatial consistency remain limitations. (source)

Use Cases

Genie 4 generates interactive worlds with native music, ambient sound and dialogue. Natural-language edits can change a scene while it runs. Gemini 3.8 Flash observes generated frames, decomposes goals and checks progress; the post also describes experiments combining Genie 4 with Gemini 4 Argon for robotics simulation. These are vendor-described capabilities, not independently validated physical predictions. (source)

Creative testers use it for previsualization, varying camera angles, lighting, dialogue and action. Google says generated video and audio include SynthID. (source)

Compared To

Within World Models, the release combines sensory generation with an external reasoning loop. Its relation to GWM Worlds 2 is a shared interactive-simulation use case; this launch provides no matched comparison establishing which is more reliable. (source)

Sources

• Google announcement and immutable capture. (source)