Google Genie 4 Timeline: What Is Verified About the World Model (Oct 27, 2026)

What Genie 4 is, on DeepMind’s pages

The DeepMind blog presents Genie 4 as a world model for real-time interactive simulation. It says the model is natively multimodal: audio such as ambient noise, music-like sound and interactive dialogue can be generated together with the visual world. The model page repeats the same pitch and hosts demo clips. Authors on the DeepMind post are Raia Hadsell, Diego Rivas and Hang Qi. The Google company blog of the same day is the shorter corporate notice and points readers back to DeepMind for demos.

DeepMind is explicit that Genie 4 builds on the Gemini family. Gemini Omni is described as anchoring simulations in real-world appearance and behavior. Other Gemini models supply reasoning that helps keep a simulation coherent as it runs. In the same post, Gemini 3.8 Flash is named as a guiding agent in the loop that observes frames, breaks goals into steps and adapts when the user intervenes. Separately, DeepMind says it has been experimenting with Gemini 4 Argon and Genie 4 for robotics research, including an Apptronik Apollo humanoid checkers example rendered as interleaved text captions into Genie 4.

The capability sections on the DeepMind blog emphasize real-time steering with natural language, more realistic physics and human motion, native audio including interactive dialogue, and multi-step events planned with Gemini in the loop. Those are DeepMind’s product claims, illustrated with demo clips on the page. Front records them as what the lab published on October 27. Photorealism and session length still need an independent measure.

Trusted-tester notes on the same blog list creative direction, historical photo immersion, interactive experience concepting and enterprise scenario generation as early use themes. They sit beside the limitations section. The lab is shipping a research preview and saying so on the same page as the demo reel.

Three access doors

Both Google and DeepMind describe a staged rollout, and the pages name three doors. Door one is a research preview for trusted testers, and DeepMind says that preview has started. Door two is Google AI Ultra subscribers in Project Genie, described as coming soon, with a waitlist linked from the DeepMind post. Door three is early-access partners of the Gemini API and Google AI Studio. A social post that says the model is out usually names none of these surfaces. The pages do not describe general availability on the consumer Gemini app or an unrestricted public API today.

The Verge (Jay Peters, Oct 27, 2026) says the reporter had not tried Genie 4, because it launches first as a research preview to trusted testers. Ars Technica the same day frames the consumer path as Google AI Ultra plans starting at $100 per month, and says pricing for Genie via the API is not yet available. The $100 figure is press framing of the Ultra subscription. It is not a Genie token price. We did not get hands-on access either. YouTube titles that say Genie 4 is here, including AI Samson’s upload, are creator framing of the demo reel.

Responsibility language on the DeepMind blog and model page says Genie 4 includes safety guardrails consistent with Google’s other models, that the preview followed internal safety evaluations, and that all generated video and audio include SynthID watermarking so AI-generated content can be identified across visual and audio modalities. DeepMind frames the trusted-tester phase as the current rollout: gather feedback before a broader release.

The company blog is the short public notice. The DeepMind blog is where capabilities, Gemini-family coupling, tester notes and limitations are written out. For a procurement or research note, cite both URLs and the dated X posts. Front’s rule for this page matches the Argon timeline: primary pages for lab claims, press labeled as press, and no invented earlier Genie generation dates. Genie stays off the Gemini language-model timeline on purpose. That timeline lists language models.

Limitations DeepMind named

DeepMind lists Genie 4 as early research. On the blog and model page it warns that visual details and object permanence can drift over longer sessions, after complex prompt edits, or when navigating a scene; that complex physical interactions, real-time audio accuracy, legible text rendering, spatial accuracy and long-term consistency remain active research areas; that continuous interaction is measured in minutes rather than extended sessions; and that real-world locations are not simulated with perfect accuracy. Front records those caveats because they are primary-source limits.

The model page phrases several of the same points as dedicated limitation headings: coherent environments over time, interaction among multiple agents, accurate real-world locations, text rendering, limited interaction duration and audio accuracy. If you are writing an internal evaluation checklist, copy those headings from deepmind.google/models/genie/. The SynthID commitment applies to generated video and audio on the pages we read. How a downstream app will show that watermark to an end user is a separate product question.

The Verge reports experiences about 2 minutes long, up from 1 minute for Genie 3. That pair of durations is press-reported. We did not time a session, and we did not find those minute counts printed on the DeepMind pages we used for the launch nodes. DeepMind’s own pages say continuous interaction is measured in minutes and list limited interaction duration among the limitation headings.

What we could not verify

• Public launch dates for Genie 1, Genie 2 or Genie 3. This page is Genie-4-centered. We did not fetch earlier DeepMind pages that might date those generations.
• A general-availability date for Project Genie consumers or for unrestricted Gemini API access. The pages say coming soon, waitlist, and early-access partners.
• Independent measurements of photorealism, physics fidelity, audio quality or session length. Demo clips and qualitative claims are DeepMind’s. The Verge’s 2-minute and 1-minute figures stay press-reported.
• A Genie token price. The primary pages we read do not publish one. Ars Technica’s $100 per month figure is Ultra-plan framing, not a Genie SKU.
• Hands-on use by FrontierDot. We did not receive trusted-tester, Ultra, or API-partner access.

Our take

Genie 4’s verified story on Oct 27 is a gated research preview with native audio and Gemini in the loop. Three doors are open in the text we read. Trusted testers have the research preview. Google AI Ultra subscribers have a Project Genie waitlist marked coming soon. Early-access partners have a Gemini API and AI Studio path. Keep this page as a live timeline. A Project Genie consumer open, an API general-availability note, or a model-card update each get a dated node when a primary URL exists.

The October 27 X posts from @GoogleDeepMind are time stamps for the announcement thread. They do not replace the blog text. When a later surface opens, expect a new primary URL or changelog line. Until then, the status line is trusted tester, Ultra waitlist, and API partner early access. We did not get hands-on, which matches what The Verge’s reporter wrote about his own access.

Related reading on Front: the Google Argon timeline and the Gemini model release timeline cover the Gemini side of the stack this world model says it builds on. Genie is out of scope for that language-model timeline on purpose. Every post is on the Front blog index.

Keep reading on Front

Every post, newest first, is on the Front blog index.

• Google Argon timeline: the dated record of Google’s Gemini 4 Argon launch.
• Gemini model release timeline: every Gemini version from 1.0 to Gemini 4 Argon.
• What you can do with Google Argon today: practical steps for Gemini 4 Argon.
• AI model releases in 2026: the 2026 nodes from the vendor timelines, in date order.

FAQ

Update log

• October 29, 2026 — Deepened Oct 29 with the three access doors, The Verge’s press-reported 2-minute versus 1-minute session note, and Ars Technica’s $100 per month Ultra framing. Still no verified Genie token price, GA date, or Genie 1–3 launch dates.
• October 29, 2026 — First published. Checked DeepMind’s Genie 4 blog and model page, Google’s Introducing Genie 4 post, and @GoogleDeepMind on X through Oct 29, 2026 (UTC+8).

Sources

• Google DeepMind blog: Genie 4 (Oct 27, 2026)
• Google blog: Introducing Genie 4
• Google DeepMind: Genie model page
• Google DeepMind on X: Genie 4 (Oct 27)
• Google DeepMind on X: Genie 4 thread reply (Oct 27)
• The Verge: Google Genie 4 (Jay Peters, Oct 27, 2026)
• Ars Technica: Genie 4 (Ryan Whitwam, Oct 27, 2026)
• YouTube: AI Samson, Genie 4 title (creator claim)