
Five to Try · Issue #10 · Thursday 29 October 2026
G'day, and happy Thursday. Five releases from the last day and a half, why each one matters, and how to try it. Google DeepMind showed off Genie 4, an AI that builds a world you can walk around in, complete with sound and people who talk back. There's also a family of tiny AI models built to run on your phone, Replit letting non-coders send fixes to their company's app, a big kung fu action game that unlocks at 1pm today, and a free AI first-aid and survival guide that works with the internet off. About a five-minute read.
1. The big one: Genie 4, Google's AI that builds a world you can step into, with sound

Image: Google DeepMind
Genie 4 is Google DeepMind's new "world model". Instead of making a video you just watch, it generates a 3D scene in real time that you move through with your keyboard, like a video game nobody had to build. You can type to change things as you go: make it rain, move the camera or drop in a new person or object. This version adds sound for the first time, from wind in the trees to characters who talk and react to you, along with more realistic physics and far more natural-looking people. DeepMind says it's still early: scenes stay consistent over short sessions but details can drift the longer one runs, and everything it makes carries Google's invisible SynthID watermark.
Why it matters: this is the tech that could one day turn a sentence into a playable game, a training ground for robots or a walk-through of somewhere you've never been. For now, all the demos are Google's own.
TRY IT → not yet for most of us. It's with a small group of trusted testers, and it's "coming soon" to Google AI Ultra subscribers (Google's top AI plan) in Project Genie, and to some developers through the Gemini API. Joining the waitlist is free, but Google hasn't said when it opens or which countries come first. Google DeepMind: Genie 4 · Genie 4 waitlist
2. Run it yourself: Liquid AI's LFM3, tiny AI models built for your phone and laptop

Image: Liquid AI (Hugging Face model card)
Most AI news is about giant models in giant data centres. Liquid AI has gone the other way. LFM3 is a family of six small models, each around 2 billion parameters, for text, images, audio and all three at once, plus two built for making decisions. The text model squeezes into about 1 GB of memory, can read very long documents and, on Liquid's own tests, beats its older, bigger model while being about a third smaller. It's built to run on phones, laptops and even a Raspberry Pi, and to be trained further on your own data using a single graphics card.
Why it matters: small, fast models that run on your own device are cheaper, quicker and more private than sending everything to the cloud. They're the kind of AI that ends up quietly built into apps you already use. The speed charts are Liquid's own.
TRY IT → try them in your browser in Liquid's online playground, or download the "open-weight" models free from Hugging Face to run yourself. The licence is free for personal use, research and businesses under US$10 million (about A$14.2 million) a year in revenue; bigger companies can't use them commercially under it. Liquid AI: Introducing LFM3 · Liquid Playground · Hugging Face: LFM3-1.7B
3. Vibe coding: Replit's Cloud Agents let non-coders send fixes to the company app

Image: Replit
Every workplace has someone who knows exactly what's wrong with the company's app: the support lead fielding the same confused customers, or the designer who's spotted wonky spacing for months. Usually they file a ticket and wait. Replit's Cloud Agents let them describe the fix in plain English instead. The AI finds the right code, makes the change and shows a live preview to tweak. It then opens your app in a browser, clicks through the change like a customer would, and sends it to the engineers as a proper request with screenshots and a video of it working. Engineers still decide what ships. An admin sets up the app's environment once and everyone shares it.
Why it matters: "vibe coding" has mostly been about building new things from scratch. This is about letting the people closest to customers improve the real product, without a developer holding their hand.
TRY IT → on Replit's Pro plan, US$100 (about A$142) a month, or US$90 (about A$128) a month billed yearly, and Enterprise (custom pricing). Changes go to your team as GitHub pull requests or GitLab merge requests. The feature that turns review comments into standing instructions for the AI is still in early access. Replit: Cloud Agents · Replit pricing
4. Games: Phantom Blade Zero, a kung fu action epic that unlocks at 1pm today

Image: S-GAME (Steam)
Phantom Blade Zero, from Chinese studio S-GAME, mixes old-school wuxia (Chinese martial arts) films with dark "kungfupunk" fantasy. You play Soul, an assassin framed for murdering his master and left with just 66 days to live, who sets out to unravel the conspiracy. The fighting is the star: more than 30 main weapons, plus 20-odd special "Phantom Edges" taken from beaten enemies, with combos the studio says are easy to learn and hard to master. It unlocks at the same moment worldwide, which is 1pm AEDT today. If you've pre-ordered, you can download it ahead of time.
Why it matters: it's one of the year's most anticipated action games, it's been near the top of Steam's global best-seller chart before launch, and it's a big-budget release from a Chinese studio.
TRY IT → A$84.95 on Steam for PC (A$99.95 for the Digital Deluxe Edition) or A$89.95 on the PlayStation Store for PS5. On PC you'll need at least 16 GB of memory, a graphics card like a GTX 1660 and an SSD. The boxed PS5 version isn't due until 2027. Steam: Phantom Blade Zero · PlayStation Store (AU)
5. Wildcard: Survival LLM, a free AI first-aid and survival guide that works with the internet off

Image: Survival LLM (brangus.io)
When the power's out and the mobile network's down is exactly when you can't search for how to treat a burn or keep the fridge food safe. Survival LLM is a free app that puts an AI assistant and a 1.5 million-article library on your own computer, so it works with no internet at all. The library covers first aid and medical references, emergency guides, survival manuals, how-to and repair guides, and about 930,000 Wikipedia articles. Every answer lists the articles it used, and you can click through to the exact passage. It picks an AI model to suit your computer's memory, and you can add your own files too, like your generator's manual or the family emergency plan.
Why it matters: with bushfire and storm season here, an offline reference is handy to have. Just note the sources lean American (the example answer says to call 911; here it's 000), and in a real emergency, call 000 first.
TRY IT → free and open source, with no account. For Apple Silicon Macs (M1 or newer) on macOS 14 or later, and 64-bit Windows 10 or 11, with at least 16 GB of memory. Set it up while you're online: the AI model is a 6 GB to 80 GB download depending on your computer, plus up to 9 GB for the library, and it takes 20 minutes to about 2½ hours. Smaller models make more mistakes, so check the sources. Survival LLM · GitHub: survival-llm · Show HN post
Know someone who'd like this? Forward it on. They can sign up free at https://fivetotry.beehiiv.com/subscribe
Five to Try is curated with AI help and checked by a human. Hit reply and tell us what you tried.
Want your product in front of curious Aussie tech readers? Sponsor Five to Try at [email protected].
