Gemini API: Genie 4 Reaches General Availability for Real-Time Interactive World Generation
Google made Genie 4 (models/genie-4) generally available in the Gemini API through the Interactions API. The model streams steerable, real-time interactive worlds at 24 fps with stereo audio, seeded from text, an image, or both. Paid-tier pricing starts at $1.50 per 1M input tokens, with video output billed at $17.50 per 1M tokens (about $0.10 per second).
Key Takeaways
- Genie 4 is now generally available in the Gemini API as
models/genie-4, moving past its earlier limited access. - It streams steerable 24 fps, 16:9 video with 48 kHz stereo audio in real time over WebRTC and WebSockets.
- Worlds can be seeded from text, a 16:9 reference image, or both, giving developers control over the starting point.
- Users steer with WASD or arrow keys plus mid-session text events, so a world can change as it is explored.
- Paid-tier pricing is $1.50 per 1M input tokens and $9.00 per 1M text output tokens.
- Video output is the main cost, at $17.50 per 1M tokens (about $0.10 per second, audio included), so session length drives spend.
Sources & Mentions
1 external resource covering this update
Real-Time Worlds From a Prompt
Google moved Genie 4 to general availability in the Gemini API on November 19, 2026. The model is exposed as models/genie-4 and is available through the Interactions API (v1beta). Instead of returning a fixed clip, Genie 4 generates an interactive world that a user can explore and steer while it streams.
What Genie 4 Streams
Genie 4 streams 16:9 video at 24 frames per second together with 48 kHz stereo audio. The stream is delivered over WebRTC and WebSockets, which suits live, low-latency sessions. Because audio is generated alongside video, no separate sound pipeline is needed for a basic experience.
Seeding a World
A session can begin from a text description, from a 16:9 reference image, or from both together. This lets developers anchor the look of a world with an image while describing behavior and setting in text.
Steering in Real Time
Users steer with WASD or arrow keys, and developers can send text events in the middle of a session to change what happens in the world. Movement and prompt-based changes can be combined, so a world can be explored and reshaped at the same time.
Pricing
On the paid tier, Genie 4 costs $1.50 per 1M input tokens and $9.00 per 1M text output tokens. Video output is billed at $17.50 per 1M tokens, which Google estimates at about $0.10 per second, with stereo audio included. Since video output dominates the cost, session length is the main driver of spend.
Who It Is For
Genie 4 targets prototypes and products built around interactive media, such as games, simulations, training environments, and creative tools. Developers who already use the Interactions API can reach it through the same surface they use for other Gemini models.