G'day, and happy Friday. Five releases from the last day, why each one matters, and how to try it. Google unveiled Gemini 4 Carbon, its strongest AI model yet, and is handing it to security experts before anyone else. There's also a free NVIDIA model that writes down who said what in your meetings, a set of genuinely fun browser games built by simply asking an AI, a much-loved detective game remastered and out today, and a free way to zoom from a galaxy down to an atom in the Gemini app. About a five-minute read.

Just two weeks after Gemini 4 Argon, Google has announced Gemini 4 Carbon, which it calls its strongest model yet. It's built for long, hard jobs: Google says its own teams have watched it work for days on end, pushing past dead ends to track down and fix subtle bugs in huge codebases. Its standout skill is cybersecurity. On Google's figures it leads outside tests for finding software flaws, checking whether they can actually be exploited and writing the fix. Google also says it's more honest: it makes about half as many unsupported claims as Argon and flags what it's unsure about rather than bluffing.

Why it matters: an AI that's this good at finding security holes is a gift to defenders and a worry if attackers get it, which is why Google is releasing it slowly. All the benchmark results are Google's own.

TRY IT → not yet for most of us. This week it goes to a small group of "cyber defenders", then to more trusted partners through Google's Fairwind Program, with developers, businesses and everyday users to follow "as soon as possible" (no date or countries given). When it reaches developers it'll cost US$2 (about A$2.84) per million "tokens" (small chunks of text) going in and US$10 (about A$14.20) per million coming out, rising to US$4 (about A$5.68) and US$20 (about A$28.40) after an introductory period. Google: Gemini 4 Carbon

Transcription apps can catch every word and still leave you guessing who said what. NVIDIA's new Nemotron 3.5 Transcribe does both in one model. It turns recordings of up to an hour into text in 40 languages, labels up to eight speakers, translates between English and 31 other languages, and can answer questions like "What decisions were made in this meeting?" You steer it in plain English, for example "label each speaker" or "spell these medical terms correctly", and it can add a timestamp to every word. NVIDIA says it's the most accurate open model on Hugging Face's English transcription leaderboard, and it's free to use, change and build into commercial products.

Why it matters: meeting notes, subtitles and dictation are where most people actually use AI, and a free, top-ranked model means cheaper and more private transcription tools are on the way. NVIDIA still says to spot-check names and speaker changes.

TRY IT → free. Try NVIDIA's online demo in your browser, or download the "open-weight" model from Hugging Face under the free OpenMDW licence. Running it yourself needs serious hardware for now: NVIDIA's guide uses an 80 GB data-centre graphics card, and an RTX 5090 is the only home card on its list. Smaller "quantised" versions for NVIDIA's desktop AI computers are "coming soon". NVIDIA on Hugging Face: Nemotron 3.5 Transcribe · Live demo · Hugging Face: the model

Simon Willison, a well-known developer and AI blogger, reckons October 2026 is a turning point: AI can now build computer games that are genuinely fun, not just games that look like games. Over a few weeks he "vibe coded" four (he never looked at the code) and shared them all. Tree Appreciator, a calm spot-the-difference game, came from a single prompt. Emberwick is a fantasy village adventure with a 15-minute story. Battle Drones is a multiplayer robot strategy game, and Quarter Sun Cove is a cosy pumpkin-farming game full of animal townsfolk. His favourite prompt when a game falls flat? "Make it a LOT more fun."

Why it matters: making a game used to take a team months. Now anyone with an idea can get something playable in an afternoon, which is exciting for some and unsettling for game artists and developers, as Simon acknowledges.

TRY IT → free. All four games run in your web browser, including on your phone. To make your own, Simon suggests starting with "Build a mobile-friendly three.js game where..." in Claude or ChatGPT; he mostly used Anthropic's Claude Opus 5.5. Simon Willison: The October inflection point for vibe game development · Play Tree Appreciator · Play Emberwick

The Wolf Among Us puts you in the shoes of Bigby Wolf, the reformed Big Bad Wolf, now the sheriff of a hidden New York neighbourhood where fairy-tale characters live in secret. Someone is committing brutal murders, and how you question suspects and what you choose to do changes who trusts you and how the story plays out. This remaster of Telltale's much-loved 2013 game has sharper visuals, better animation and audio, and a new Noir Mode that switches the whole thing to black and white at any time. It includes all five episodes, five playable lost scenes, an art gallery and over an hour of behind-the-scenes interviews.

Why it matters: it's one of the best-loved story games of the last decade, and the remaster is a cheap, easy way in if you missed it. It's for mature players, with strong violence, language and themes.

TRY IT → out now on PC (Steam and Epic), PS5 and Xbox One, and playable on Xbox Series X|S; it unlocked at midnight AEDT. It's US$24.99 (about A$35), with 20% off for new players until 13 November (on PS5 that's PS Plus members only) and up to 50% off if you own the original digitally on the same store. Australian store prices weren't showing before launch, so check the store. The Switch version is coming later. Steam: The Wolf Among Us Remastered · PlayStation Store (AU) · Telltale: launch discounts

Some things are easier to understand when you can poke at them. The Gemini app now has a library of "immersive explorations": interactive 3D scenes you explore layer by layer, from deep space down to the microscopic. You might trace the trade routes of the Silk Road, peer inside a jet engine or study the physics of lightning, then tap a label to go deeper, all the way down to molecules and atoms. Google says spoken narration is coming soon, and you'll soon be able to open the scenes right inside a chat with Gemini.

Why it matters: it's a glimpse of AI helping people learn by exploring, not just by reading a wall of text. Handy for curious kids, students or anyone who's ever wondered what's inside a jet engine.

TRY IT → free in the Gemini app's student hub: click "Try immersive explorations". Google says it's available to all users, in English and five other languages. Google: Immersive explorations in the Gemini app · Try immersive explorations

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].