Genie 4: What It Is, Availability, Specs and Who Should Use It
Last verified: October 28, 2026 (Asia/Shanghai). Facts below come from Google DeepMind's Genie 4 blog and model page, the Google Keyword blog post, Google DeepMind / Google / Demis Hassabis posts on X, TechCrunch, The Verge, Ars Technica, and the Hacker News thread linked in Sources. DeepMind has published no Genie 4 parameter count, model card, or Genie-4-specific API price. Demo clips are vendor-selected recordings of real-time sessions. We have not tested Genie 4 hands-on; independent press reviewers (e.g. The Verge) also report they have not used it yet.
Genie 4 is Google DeepMind's newest real-time interactive world model, announced on October 27, 2026. On the official blog, DeepMind describes it as unlocking "unprecedented realism" with native multimodal audio—ambient sound, music, and interactive dialogue generated with the world. @GoogleDeepMind called it "our most realistic and interactive world model yet." Below: what Genie 4 is, who can access it, what DeepMind claims vs. what is still unknown, how early reactions broke down, and how it compares with Genie 3 and related Google models such as Gemini 4 Argon.
| At a glance | Genie 4 |
|---|---|
| Maker | Google DeepMind |
| What it is | Real-time interactive world model (simulate explorable scenes from prompts / images) |
| Announced | October 27, 2026 (DeepMind blog; Google Keyword) |
| Status | Research preview for trusted testers; broader rollout "coming soon" |
| Consumer path | Google AI Ultra subscribers via Project Genie (waitlist) |
| Developer path | Early-access partners of the Gemini API and Google AI Studio (same blog) |
| Native audio | Yes (first DeepMind real-time world model with native audio, per DeepMind) |
| Session length (press / official) | Press: ~2 minutes (up from ~1 minute for Genie 3); official: "a few minutes" of continuous interaction |
| Genie-4-specific API price | Not published |
| Official numeric benchmarks | None published (capability demos only) |
| Watermark | SynthID on generated video and audio (DeepMind blog) |
Sources: DeepMind Genie 4 blog, Genie model page, The Verge, TechCrunch, Ars Technica.
Practical rule: treat Genie 4 as a research preview and a waitlist product—not a publicly priced API you can call today.
What Genie 4 is
Genie 4 is a world model: it generates an interactive simulated environment frame by frame as you navigate (keyboard / camera) and steer with natural-language prompts. DeepMind's pitch is that you "set an event in motion, and Genie 4 simulates what happens next: how the world responds, how it sounds, and how the people in it move, react, and speak" (blog).
DeepMind says Genie 4 is natively multimodal and builds on the Gemini family: Gemini Omni "anchors the simulations in real-world appearance and behavior," while other Gemini models supply reasoning for coherence (blog; Google Keyword). For multi-step events, DeepMind describes Gemini 3.8 Flash as a guiding agent in the loop that observes frames, breaks goals into steps, and adapts when the user intervenes (blog; also summarized by The Verge).
DeepMind groups four capability themes on the launch posts and blog:
- Real-time steering — change weather, camera, characters, or objects mid-simulation with text prompts.
- Physics and human action — vendor claims of stronger aerodynamics, collisions, and more natural people.
- Native audio — ambient noise, music, and interactive dialogue generated with the world (DeepMind: first real-time Genie with native audio).
- Gemini-guided complex events — including robotics-oriented experiments pairing Genie 4 with Gemini 4 Argon for multi-step task plans rendered live (blog; @Google).
Trusted-tester use cases DeepMind cites include film pre-visualization / ideation, turning archival photos into explorable historical scenes, game / interactive concept prototyping, and enterprise "zero-to-one" scenario generation (blog; TechCrunch).
Third-party press still from Ars Technica's Genie 4 coverage (credit: Google footage via Ars); may be slow or blocked in some regions.
Availability, pricing and API
As of October 28, 2026:
- Research preview for a limited group of trusted testers (DeepMind blog; Google Keyword).
- Coming soon to Google AI Ultra subscribers in Project Genie, and to early-access partners of the Gemini API and Google AI Studio (same sources; Demis Hassabis: "wider access coming soon").
- DeepMind points readers to a Project Genie waitlist from the blog (forms.gle waitlist linked from the DeepMind post; also listed in the official YouTube description).
- No Genie-4-specific API price, rate card, or public endpoint name appears on the DeepMind Genie page or blog. Ars Technica reports developer access via Gemini API / AI Studio with pricing "isn't yet available," and that consumer access will phase through Google's AI Ultra plans, "which start at $100 per month"—that is a statement about the Ultra subscription, not a published Genie 4 meter. Treat the $100 figure as press reporting about Ultra, and re-check Google's own plan pages before budgeting.
Practical rule: join the Project Genie waitlist or watch AI Ultra / Gemini API early-access notes; do not assume a public
$/minuteGenie 4 SKU exists yet.
Specs DeepMind has (and has not) published
| Claim | Source | Notes |
|---|---|---|
| Real-time interactive world simulation | DeepMind blog, model page | Vendor description |
| Native audio (ambient, music, dialogue) | Same | Vendor: first real-time Genie with native audio |
| Natural-language mid-session steering | Same | Weather, camera, characters, objects |
| Gemini Omni + other Gemini models for appearance / reasoning | Same; Google Keyword | Vendor architecture framing |
| Gemini 3.8 Flash as in-loop guiding agent | DeepMind blog; The Verge | Vendor claim |
| Pairing with Gemini 4 Argon for robotics sims | DeepMind blog | Research experiment; see our Gemini 4 Argon guide |
| Continuous interaction length | Official: "a few minutes" (model page); press: ~2 min vs Genie 3 ~1 min (The Verge, TechCrunch) | Not a formal SLA |
| SynthID watermarking on video + audio | DeepMind blog | Vendor responsibility claim |
| Parameter count / training data / FLOPs | — | Not published |
| Genie-4-specific API price / quotas | — | Not published |
| Public leaderboard / numeric eval suite | — | Not published |
Documented limitations (official)
DeepMind's model page and blog list, among others:
- Visual detail / object permanence drift over longer runs or after complex prompt edits
- Multi-agent interaction still a research challenge
- Real-world locations not simulated with perfect accuracy
- Legible text often only when present in the input description
- Limited continuous interaction duration ("a few minutes")
- Real-time audio accuracy still improving
Official benchmarks
None published as numeric tables. Genie 4's launch is demo- and capability-led: DeepMind shows curated real-time interaction recordings with keyboard navigation and text steering, not a SWE-bench-style scorecard. Any third-party chart that assigns Genie 4 a leaderboard number without citing DeepMind or a named independent lab is unverified—do not trust it.
Vendor capability claims (photorealism, physics, dialogue) should be read as marketing / research framing, not as independent measurements. For comparison, shipped language models on this site often do publish vendor tables—see Gemini 4 Argon, Kimi K3.1, or DeepSeek V4.1-Flash-1017—Genie 4 is not in that category yet.
Real-world reactions (demos and discussion only)
Almost nobody outside Google's trusted-tester cohort has hands-on access. Public reaction is about DeepMind's videos and blog wording.
Official / founder posts (verbatim excerpts):
- Google DeepMind: "This is Genie 4, our most realistic and interactive world model yet. Explore dynamic environments in real time to see how the scene responds, sounds, and how the people in it move, react, and speak." (X)
- Google DeepMind follow-up (full note): lists photorealism via Gemini Omni, advanced physics, real-time controls, native audio, and Gemini reasoning for multi-step coherence (X).
- Demis Hassabis: "Very excited to share Genie 4, our incredible interactive world model, bringing together real-time physics, reasoning, and audio! Simulations within simulations, makes you wonder just how deep the rabbit hole goes..." (X)
- Co-author / engineer Diego Rivas: "We started rolling Genie 4 as a research preview to trusted testers, and will be releasing it soon to Project Genie, the Gemini API and AI studio." (X)
Press: TechCrunch highlights native audio, free-form mid-sim edits, ~2-minute consistency (double Genie 3 per their reporting), Gemini-as-director, and filmmaking prototyping interest from testers. The Verge lists four main updates (realism, real-time NL updates, audio, Gemini 3.8 Flash) and notes Jay Peters has not tried Genie 4 himself. Ars Technica stresses AGI / robotics framing, Ultra-gated consumer access, and unpublished developer pricing.
Hacker News (thread, 749 points at fetch time) was active and often skeptical:
- Controllability: sodacanner wrote that Genie 4 "almost looks worse than Genie 3 in terms of player control" and questioned whether demos respond to inputs or overlay widgets on generated video (comment).
- Physics: mlmonkey argued that in the marble-bust-on-watermelon clip, "the bust should've toppled over, if the physics were properly modeled" (comment).
- Product fit: several comments framed Genie as closer to video generation / previsualization than a shippable game engine (e.g. whether assets export to Unity/Unreal).
- Positive notes: TGower called the stained-glass landscape video "pretty cool" (comment).
Preview frame from X/Twitter CDN (DeepMind announcement video); may be slow or blocked in some regions.
Watch on YouTube: "Genie 4: A new generation of world simulation" by Google DeepMind
Official Google DeepMind channel video. AI Takeaway did not make this video and is not affiliated with Google or DeepMind.
No time to watch? AI Takeaway is a tool that works on any YouTube video, including official launch clips like this one. Paste the link to get key points in minutes, jump to original timestamps, and ask follow-ups with answers that cite the video. Browse more AI and tech interviews in the Hub.
What launch coverage and reviews add
Use demos critically: timestamp-verifiable details in the recording beat adjectives like "unprecedented." Cross-check product claims against the DeepMind blog and model page. HN and press skepticism about control fidelity and physics is useful precisely because DeepMind's own limitations section already admits drift, short sessions, and imperfect audio/text.
Who should care about Genie 4
- Filmmakers / creative directors exploring pre-vis: DeepMind and TechCrunch both emphasize rapid ideation and live "takes" with camera / lighting / dialogue changes.
- Game / interactive prototype teams: useful for style and mechanic sketches before engine work—not a replacement for a deterministic shipped game loop.
- Robotics and agent researchers: especially teams already following Gemini Robotics / Gemini 4 Argon simulation experiments.
- Google AI Ultra / Project Genie users on the waitlist: the named consumer path.
- Probably not for production owners this week: no public Genie 4 price, SLA, or documented API SKU. If you need a callable language or coding model now, look at shipped guides such as Gemini 4 Argon, GPT-6.1 Sol vs GPT-6 Astra, Kimi K3.1, or DeepSeek V4.1-Flash-1017.
How it compares (briefly)
| Criterion | Genie 4 | Genie 3 (prior) |
|---|---|---|
| Role | Real-time interactive world model | Prior DeepMind world model / Project Genie base |
| Native audio | Yes (vendor) | No native audio in Genie 4's "first" claim framing |
| Mid-sim natural-language steering | Emphasized | Supported in earlier Project Genie (press) |
| Session length | ~2 min (press) / "few minutes" (official) | ~1 min (press comparison) |
| Gemini coupling | Gemini Omni + 3.8 Flash (+ Argon experiments) | Less tightly described as hybrid in Genie 4 launch materials |
| Access now | Trusted testers + waitlist | Limited Project Genie experience historically |
TechCrunch also notes a crowded world-model field (e.g. Runway GWM Worlds 2; industry M&A around World Labs / Decart / Third Dimension)—those are peers in category, not drop-in substitutes with published head-to-head metrics. For Google ecosystem context beyond world models, see Muse Wonder; for speech frontiers outside Google, see Dessa Voice AI and Grok Voice 3.0.
Practical rule: pick Genie 4 to study interactive world simulation and pre-vis research; pick a documented language/coding API when you need production tokens this quarter.
Getting-started checklist
- Read the DeepMind Genie 4 announcement and the Genie model page—including the limitations section.
- Watch the official DeepMind YouTube overview (or the Google for Developers intro).
- Join the Project Genie waitlist linked from DeepMind's blog if you want consumer Ultra-path access.
- If you are an early-access Gemini API / AI Studio partner, watch Google's partner channels for Genie 4 availability—no public price list yet.
- Skim the HN thread for controllability and physics skepticism.
- Re-check DeepMind's pages before assuming exports to Unity/Unreal, long sessions, or a public
$/minAPI—none of those were published as of this verification date.
Keep up without the watch time
World-model and frontier launches are moving weekly. If you would rather read than watch long demos, the AI Takeaway video summary hub has free curated summaries with timestamp jumps, for example:
- Muse Wonder: Zuckerberg and Wang's $100B AI
- Gavin Baker on Orbital Compute, TSMC, and Frontier Models
- Sam Altman on AGI, Compute, and Human Agency
Install the AI Takeaway Chrome extension for key points on YouTube, or submit any YouTube video on aitakeaway.net. More guides live on the articles index; keep checking the Hub as new coverage lands. Related posts: Gemini 4 Argon, Muse Wonder, GPT-6.1 Sol vs GPT-6 Astra, DeepSeek V4.1-Flash-1017, Kimi K3.1, Dessa Voice AI, Grok Voice 3.0.
FAQ
What is Genie 4?
Genie 4 is Google DeepMind's real-time interactive world model announced October 27, 2026. It generates explorable simulated environments with native audio and natural-language steering, building on Gemini models for appearance and coherence (DeepMind blog).
Can I use Genie 4 today?
Only if you are in DeepMind's trusted-tester research preview. Broader access is described as coming soon for Google AI Ultra users in Project Genie and early Gemini API / AI Studio partners (Google Keyword). Everyone else can join the Project Genie waitlist linked from the DeepMind blog.
How much does Genie 4 cost?
DeepMind has not published a Genie-4-specific API price. Consumer access is planned through Google AI Ultra / Project Genie; Ars Technica notes Ultra plans starting at $100 per month, which is about the subscription tier—not a confirmed Genie 4 meter (Ars Technica).
How long can a Genie 4 session run?
DeepMind's model page says the model can support "a few minutes of continuous interaction, rather than extended sessions." The Verge and TechCrunch report about two minutes, up from about one minute for Genie 3—press comparisons, not an SLA.
Does Genie 4 replace game engines?
No published path exports a Genie 4 session as a Unity/Unreal/Godot project. DeepMind positions it for simulation, creative prototyping, and agent/robotics research; HN commenters widely treated AAA replacement claims as speculative.
How does Genie 4 relate to Gemini 4 Argon?
DeepMind shows research demos where Gemini 4 Argon plans multi-step robotics-style tasks and streams text captions for Genie 4 to render in real time (DeepMind blog). Argon is a language/reasoning model with public API pricing; Genie 4 is a world simulator still in preview—see our Gemini 4 Argon guide.
Sources
- Google DeepMind, Genie 4: A new generation of world simulation (Oct 27, 2026)
- Google DeepMind, Genie 4 model page
- Google Keyword, Introducing Genie 4 (Oct 27, 2026)
- Google DeepMind on X, announcement and what's new
- Demis Hassabis on X, Genie 4 and waitlist note
- Google on X, introducing Genie 4 thread
- Diego Rivas on X, launch note
- TechCrunch, Google's Genie 4 world model adds audio… (Oct 27, 2026)
- The Verge, Google's next Genie AI model… (Oct 27, 2026)
- Ars Technica, DeepMind unveils Genie 4… (Oct 27, 2026)
- Hacker News, Genie 4
- YouTube (official): Google DeepMind – Genie 4; Google for Developers – Introducing Genie 4
- Related AI Takeaway guides: Gemini 4 Argon, Muse Wonder, GPT-6.1 Sol vs GPT-6 Astra, DeepSeek V4.1-Flash-1017, Kimi K3.1, Dessa Voice AI, Grok Voice 3.0