
Genie 4 Is Live: What Google AI Ultra Access Actually Buys You
- OrcaNEWOrcaCyber: Hy4 Flash Abliterated / Uncensored2026-11-18$0.07 / $0.28 per 1M tokens · 184 tok/s
- tencentNEWTencent: Hy4 Flash2026-11-16$0.02 / $0.07 per 1M tokens · 228 tok/s
- GoogleNEWGoogle: Gemini 4 Carbon2026-11-10$2.00 / $10.00 per 1M tokens · 87 tok/s
- AnthropicNEWAnthropic: Claude Fable 5.22026-11-09$8.00 / $40.00 per 1M tokens · 96 tok/s
- orcaNEWOrca: OrcaC2 1.02026-11-079 tok/s
- OrcaOrcaCyber: GLM 5.3 Abliterated / Uncensored2026-11-04$3.00 / $5.00 per 1M tokens · 90 tok/s
- OrcaOrcaCyber: Hy4 Abliterated / Uncensored2026-11-04$0.80 / $1.60 per 1M tokens · 131 tok/s
- OrcaOrcaCyber: Kimi K3.1 Pro Abliterated / Uncensored2026-11-04$2.00 / $5.00 per 1M tokens · 77 tok/s
- tencentTencent: Hy42026-11-04$0.83 / $2.50 per 1M tokens · 94 tok/s
- OrcaOrcaCyber: MiniMax M3.1 Flash Abliterated / Uncensored2026-10-29$0.80 / $1.60 per 1M tokens · 114 tok/s
- OrcaOrcaCyber: GLM 5.3 Flash Abliterated / Uncensored2026-10-27$0.80 / $1.60 per 1M tokens · 69 tok/s
- OrcaOrcaCyber: Qwen3.8 Flash Next Abliterated / Uncensored2026-10-26$0.80 / $1.60 per 1M tokens · 94 tok/s
- OrcaOrcaCyber: MiMo V2.6 Flash Abliterated / Uncensored2026-10-26$0.80 / $1.60 per 1M tokens · 41 tok/s
- MoonshotMoonshotAI: Kimi K3.1 Pro2026-10-2447Intelligence
- OrcaOrcaCyber: DeepSeek V4.1 Flash Abliterated / Uncensored2026-10-23$0.80 / $1.60 per 1M tokens · 208 tok/s
- minimaxMiniMax: M3.1 Flash2026-10-1929Intelligence
- OrcaOrca: OrcaCyber Ultra 1.02026-10-17$5.00 / $10.00 per 1M tokens · 111 tok/s
- deepseekDeepSeek: DeepSeek V4.1 Flash 10172026-10-1733Intelligence
- DeepSeekDeepSeek: DeepSeek V4.1 Flash 09102026-10-17$0.15 / $0.60 per 1M tokens · 311 tok/s
- DeepSeekDeepSeek: DeepSeek V4.1 Flash2026-10-1740Intelligence
Genie 4 stopped being a waitlist on November 18, 2026. That is the day Google DeepMind began rolling its most realistic world model out to eligible Google AI Ultra subscribers inside Project Genie, and the day the release stopped being a demo you read about and became a product you can pay for. The awkward part is that the same update also made Genie 3 better in the one dimension that decides most practical work: Google's older world model now runs three-minute sessions with world chaining and 360-degree panorama grounding, while the Genie 4 model card still describes a system that supports "around a minute of continuous interaction." If you are deciding whether to spend Ultra money this week, that inversion — not the photorealism — is the thing to understand. And if you are building around it, the only half of the stack you can actually call today is the reasoning half: Gemini 3.8 Flash and Gemini 4 Argon are live endpoints, Genie 4 is not an endpoint anywhere.
Two Dates, Not One, and Why the Gap Matters
Google announced Genie 4 on October 27, 2026, to trusted testers and a waitlist. Nobody outside that tester pool could reach it. The November 18 post is a different kind of document: it is the availability announcement, and it moves the model from "reported" to "purchasable," which is the only transition that changes what a reader can do today.
What the post actually says, in Google's own words, is narrow. Genie 4 rolls out to eligible AI Ultra subscribers. Google AI Pro members can join a waitlist. The same update ships Genie 3 upgrades, the Project Genie agent, and a community gallery. There is no general-availability date, no API model ID, no pricing per generation, and no benchmark. Three weeks of coverage have not produced a resolution figure or a frame rate either.
So the honest summary of November 18 is this: the model is real, the door is open, and the door is a $99.99 monthly subscription.
The Price, and What It Is a Price For
Google AI Ultra starts at $99.99 per month in the US, and the same page lists a $199.99 tier with 20x the usage limits of AI Pro rather than 5x. Both include Project Genie; the difference is quota, not capability. AI Pro at $19.99 does not get you in — it gets you a waitlist form.
• Entry price for Genie 4 — $99.99/month (Google AI Ultra, 5x AI Pro limits)
• Higher tier — $199.99/month (20x AI Pro limits, same model access)
• Waitlist tier — Google AI Pro at $19.99/month, no access today
• Age and geography — 18+, Project Genie available in more than 140 countries per Google's own footnotes
• What you are not buying — an API. There is no per-token or per-second rate card for Genie 4 anywhere, and Google has not published one.
That last line is the part procurement teams miss. A subscription to a browser product is a seat, not a service level. You cannot put a seat behind an automated pipeline, you cannot burst it, and you cannot route around it when it degrades. The subscription buys you the ability to evaluate Genie 4 by hand — which is genuinely worth $99.99 to a studio doing pre-visualization research, and worth close to nothing to a team that needs a world model in a build.

The Genie 3 Upgrade Is the Usable Half of This Release
Buried under the Genie 4 headline is a set of Genie 3 changes that matter more for anyone doing actual work, because Genie 3 is the model with the longer session budget. Google lists four:
• Extended sessions and world chaining — up to 3 minutes per generation, up from 60 seconds, with multiple worlds linked into a longer journey
• 360-degree panorama grounding — seed a new world from a panorama image and keep the location consistent as the camera looks around
• Real-time in-world events — natural-language prompts that change the environment mid-session, not just between sessions
• Model picker — Genie 3 stays selectable alongside Genie 4, framed by Google as the choice for longer runs and panorama-seeded spaces
Read those together and Google is telling you something it will not say outright: for a three-minute continuous shot, pick Genie 3. Genie 4 wins on fidelity, physics and audio; Genie 3 wins on duration and spatial grounding. A director blocking a sequence gets more from three grounded minutes than from one beautiful one, and a sound designer gets nothing from Genie 3 at all, because the native audio is Genie 4 only.
Genie 4's Own Spec Sheet Is One Number Long
The Genie 4 model page is a study in what a vendor will and will not commit to. It claims the model is "our most realistic and interactive world model yet," describes real-time prompt edits to weather, camera, characters and action, and describes natively generated audio — ambient soundscapes, footsteps, crowds, character speech — that adapts as the scene changes. Then it lists the limitations in the same breath.
• Duration — "around a minute of continuous interaction, rather than extended sessions"
• Object permanence — visual details and object permanence "can shift or drift over time," especially after heavy editing
• Multi-agent scenes — modelling several users and agents in one shared space is called an ongoing research challenge
• Real locations — Google states the model cannot simulate real-world locations with perfect accuracy
• On-screen text — legible text usually appears only when it came from the input description
• Audio accuracy — "remains an area of ongoing research and improvement"
• Status — "an early, evolving technology"
No resolution, no frame rate, no latency figure, no parameter count. Every claim in the launch material is relative to Genie 3 or to Google's own prior output. That is not a reason to dismiss the model — native audio in a world model is a real capability change — but it is a reason to treat "realistic" as a vendor adjective rather than a measurement. Independent numbers do not exist yet, because the people with access are subscribers, and subscribers do not publish evals.
The Agent and the Gallery Are the Actual Product
Two additions in this release have nothing to do with the model weights and everything to do with why a subscription is the delivery vehicle.
The Project Genie agent takes a natural-language description and builds a world from it, coordinating Google's multimodal models to produce the initial visuals and a custom background score. In-session, Gemini models watch the screen, surface suggestion chips, and can trigger in-world events. That is the same architecture the earlier Genie 4 coverage flagged and most coverage skipped: a world model plus a planner is a different artifact from a video generator, because it answers an objective with a sequence of validated states rather than answering a prompt with a clip.
The community gallery is the distribution play. Curated worlds sit beside user-built ones, and any of them can be opened, remixed, or republished. Google commissioned two showcase pieces — "Villa Giallo" with creative technologist Will, and "The Life of Ants" with filmmaker Anand Gandhi — which is a familiar pattern: seed the gallery with professional work so the first wave of user output has something to be measured against. All generated video and audio carries SynthID watermarking, and inputs and outputs go through safety screening.
What the gallery does not settle is licensing. Google has published no commercial-use terms for worlds built in Project Genie, which is a larger blocker for a studio than the one-minute ceiling. A pre-vis vendor that cannot tell a client who owns the shot has a problem the model card does not address.

If You Are Building the Loop, Half of It Is Already an API
Here is the practical split. Genie 4 is a browser product with no endpoint. The models around it — the ones doing the planning, the verification, the suggestion chips — are ordinary LLM endpoints, and those you can call today.
Gemini 3.8 Flash is the natural fit for the watching-and-checking role: it is Google's most intelligent Flash model, built for software engineering and multi-step agentic work, with a 1,048,576-token context window and 65,536 max output tokens. On OrcaRouter it is google/gemini-3.8-flash at $0.75 per million input tokens and $3.75 per million output, with cached input reads at $0.075. Gemini 4 Argon is google/gemini-4-argon at $2.00 and $10.00 — the model you reach for when the plan itself is the hard part, as it was in Google's robotics demo where Argon decomposed tasks that Genie 4 then simulated.
Two things make that split worth setting up properly rather than hacking together. First, pricing: OrcaRouter passes provider list price through at 0% markup, so if Google cuts Gemini pricing the cut is live on your key the same day, with no repricing negotiation. Second, the architecture: a world model you cannot fail over is a single point of failure, and a planner you can is not. Automatic failover across providers, plus a routing DSL that can pin the cheap model for verification passes and the stronger one for planning, is exactly the shape a Genie-adjacent pipeline needs — because the expensive, unavailable half is the half you cannot control.
Be clear about what this is not. Genie 4 is not on our catalogue and is not on anyone's. No amount of routing makes an unreleased endpoint callable, and any article implying otherwise is selling you something Google has not shipped.

Who Should Pay, and Who Should Wait
Subscribe this week if you do pre-visualization, level prototyping, or robotics simulation research and you need to know by hand whether native audio plus a two-minute-class world is enough to change your pipeline. The $99.99 is a research budget, and the answer is genuinely not knowable from demos — audio sync over a full minute is where generated audio usually breaks, and nobody outside the tester pool has reported on it.
Wait if any of the following is true: you need an API, you need a commercial license, you need a session longer than three minutes, or your work depends on object permanence surviving an edit. Genie 3 now covers the third case better than Genie 4 does. Nothing in this release covers the first two.
The thing to watch is not another demo reel. It is three specific publications: a resolution and frame rate from anyone with access, an independent check of whether the audio stays locked to its source across the whole session, and a terms-of-use page that says who owns a generated world. Until those land, Genie 4 is a subscription you can evaluate and an API you cannot buy — and the gap between those two sentences is where every build decision this quarter gets made.
