G'day, and welcome to the week. Five releases from the last few days, why each one matters, and how to try it. Mistral has released its biggest AI model for anyone to download, and Sydney's own Canva has built an AI that edits your designs without wrecking them. There's also a slop-free YouTube that a famous AI researcher built on a Saturday morning, a rhythm platformer from India where every jump lands on the beat, and a free way to edit documents right inside Gemini. About a five-minute read, so it'll fit before the first coffee's gone cold.

Image: Mistral

French AI company Mistral has released the "weights" (the trained model itself) of Mistral Large 4, nicknamed Le Chonk, so anyone can download it, run it and change it. It's huge: about a trillion "parameters" (the internal settings an AI learns), it can read images as well as text, and it's fluent in more than 160 languages. Since a preview earlier in October, Mistral has kept training it, and says it's now the best open model for coding and for defensive cybersecurity work like finding and patching software flaws, while refusing more malicious hacking requests than any other open model it measured. It also has three thinking levels (low, medium and high), so you can trade a little quality for speed and cost.

Why it matters: a top-tier AI that businesses and governments can run on their own computers, without sending data to an American or Chinese company, is a big deal for privacy and for anyone wary of relying on one provider. The benchmark results are Mistral's own.

TRY IT → the download is free for personal and commercial use under Mistral's own licence (companies that sell access to the model as a service need a separate licence). Realistically it's a data-centre model: Hugging Face's ready-made setup uses eight NVIDIA H200 graphics cards. For the rest of us it's on Mistral's API, Mistral Studio and OpenRouter at US$0.70 (about A$1.00) per million "tokens" (small chunks of text) going in and US$2.10 (about A$2.99) per million coming out. Developers can also point Mistral's Vibe Code tool at their own projects. Mistral: Mistral Large 4 is now open weight · Hugging Face: the model

Image: Canva

Most AI image tools hand you a flat picture you can't tweak. Canva's new Brush, built by Canva Research, its 160-person AI lab, works the other way: it looks at your existing design, understands how it's put together and makes the change you ask for, from fixing one line of text to restyling every slide in a deck. Behind the scenes it works like a small agent (an AI that does a task in steps), writing and running bits of code to move things around, then checking the page before its next move. Everything it touches stays fully editable, so you can take over by hand at any time. Canva says it's up to 69% faster than its previous design AI.

Why it matters: Canva is a Sydney success story used by hundreds of millions of people, and building its own AI models (after its Craft image model a week earlier) means faster, cheaper AI help for everyday design jobs like resizing a slide into a social post.

TRY IT → Canva Pro only for now (A$20 a month or A$165 a year on Canva's Australian pricing page), with wider availability "coming soon". Open a design, go to Canva AI and choose "Brush" from the model menu. It's labelled beta. Canva: Introducing Canva Brush · Canva AI

Image: HackerTube (screenshot)

On Friday night, Andrej Karpathy, the well-known AI researcher who coined the term "vibe coding", opened YouTube and found what he called a "wall of slop". So on Saturday morning he asked Anthropic's Claude to build a better one. The result, HackerTube, is a YouTube look-alike filled only with videos that people on Hacker News (a popular tech discussion site) have recommended to each other: 6,454 videos across 150 topics, gathered from about 40,000 comments going back to 2007. It shows a frame from the middle of each video instead of the clickbait thumbnail, and its "For You" picks learn from what you watch without an account. Karpathy says it took about two hours, and he didn't write any code himself.

Why it matters: this is vibe coding in a nutshell. One person with an itch built, launched and secured a whole website in a morning, and the result is something genuinely nice to use.

TRY IT → free, no sign-up, in any web browser. Karpathy also offers the site's data to download, so you can ask your own AI helper to build your own version. He used Claude Code with Claude Opus 5.5. Karpathy: HackerTube · Watch on HackerTube

Image: Tathvamasi Studios (Steam)

In SURI you play Ajira, a fast, fearless young girl searching an island for a fruit that can save her sick mother. On this island rhythm is the law of nature: you run, jump and dash in time with the music, and missing the beat breaks your flow. It's the debut game from India's Tathvamasi Studios, four years in the making and backed by PlayStation's India Hero Project. The hand-drawn world borrows from Rajput forts, Himalayan monasteries and Kerala's Theyyam culture, and most of the soundtrack was recorded on real instruments, with the sounds changing from region to region. The team even spaced the main beats 480 milliseconds apart so your thumbs can keep up.

Why it matters: it's a fresh, beautiful world you don't often see in games, and a rhythm game that should suit both platformer fans and music lovers.

TRY IT → out now on PC (Steam) for A$29.50, or A$23.60 with 20% off until 13 November. It's also on PS5; check the PlayStation Store for the Australian price. The Steam version is Windows only and needs at least a GTX 1650 graphics card and 8 GB of RAM. Steam: SURI: The Seventh Note · PlayStation Blog: SURI launches 30 October

Image: Google

Until now, getting a document out of an AI chat usually meant copying text into Word or Google Docs and tidying it up there. Google's Gemini app can now create, preview and edit files right inside your chat: documents, PDFs, slide decks, spreadsheets, code and more. Go full screen to type changes yourself, highlight a section and ask Gemini about it, or ask it to rewrite text or redo a chart. Google's example: snap your handwritten lecture notes, ask for a PDF study guide, then fix it up without leaving the chat.

Why it matters: it cuts out the copy-paste shuffle, and it's a small step towards AI that finishes the job rather than handing you a rough draft to clean up.

TRY IT → free. Google says it's available to all users in every language the Gemini app supports. PDFs work on the web and on your phone; the other file types (Google Docs and Sheets, slide decks, plain text and code) are web only for now. Google: Edit and preview documents directly in Gemini · Open Gemini

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].