G'day, and welcome to the week. Five releases from the last few days, why each one matters, and how to try it. Google's Gemini Omni 1.2 Flash can now make 15-second videos and costs 30% less, and Thinking Machines has put Inkling 2 Small on Tinker, with open weights promised after safety testing. Cursor will build a web app, set up a database and host it from one prompt to a cloud agent, and it has also added a realtime voice call with your coding agents. There is a cheap party game too, where you and your friends redub any video clip. About a five-minute read.

Google has released Gemini Omni 1.2 Flash and made it generally available in the Gemini API and on Google Cloud. Clips can now run to 15 seconds, up from 10 seconds in Omni 1.1, and you can pick any length from 3 to 15 seconds in one-second steps. You can also edit the same clip turn by turn in conversation — swapping a product, background or aspect ratio — instead of generating the whole thing again. Google says the model holds on to reference images, short motion clips and audio cues more tightly, so faces, packaging and brand colours stay consistent across camera moves. The headline price is 30% lower than Omni 1.1: US$0.07 a second for 720p video (about A$0.10 a second, or about A$1.53 for a 15-second clip). You can draft at 360p or 720p and upscale keepers to 1080p or 4K. Outputs carry an imperceptible SynthID watermark.

Why it matters: a 15-second clip is long enough for a social ad or a product explainer, conversational edits save you regenerating the whole thing, and the lower per-second price makes it cheaper to try more takes.

TRY IT → anyone with a Google account or Google Workspace subscription can generate up to three 15-second videos a month in Google Vids at no extra cost (try it at vids.new). Developers can use Gemini Omni 1.2 Flash in the Gemini API at US$0.07 a second for 720p (about A$0.10). It is also rolling out in Flow, Google Ads and the Gemini app for Google AI Ultra subscribers. Google: Gemini Omni 1.2 Flash · Google Vids: Omni 1.2 Flash at no cost

Thinking Machines Lab has launched Inkling 2 Small, a mixture-of-experts model with 594 billion total parameters and 17 billion active. It accepts image inputs as well as text, and it lets you vary how much effort the model spends on a task. The lab says it scores 31 on the Artificial Analysis Intelligence Index, well above its earlier Inkling models, with the biggest gains in agentic coding, vision, knowledge work and reasoning. For now the weights are closed and the model runs on Tinker. Thinking Machines says it intends to publish the full weights under Apache 2.0 after external safety testing and red-teaming, which the 5 November launch post described as "next week". As of this weekend those weights were not out yet.

Why it matters: a strong, tuneable US model you can try and fine-tune today is useful even before the download lands, and an Apache 2.0 release would make it one of the more open options at this size. The scores are Thinking Machines' own, and the weights are still gated until the safety work is done.

TRY IT → try it on Tinker Playground today, or use it for inference and fine-tuning on Tinker. Serverless inference (in beta) lists Inkling 2 Small at US$0.30 per million input tokens and US$1.00 per million output tokens (about A$0.44 and A$1.45), with cached input at US$0.03 (about A$0.04). Training is US$1.50 per million tokens at 64K context or US$3.00 at 256K (about A$2.18 and A$4.36). Thinking Machines: Introducing Inkling 2 Small · Tinker pricing

Cursor has added Live Apps: you start a cloud agent and tell it you want to build an app, and it scaffolds the site, spins up a Postgres database if the project needs one, puts sign-in behind Cursor SSO and provisions hosting. When you are ready you pick a unique domain and publish. Nothing goes live until security checks pass and you explicitly hit publish. Those checks look for secrets written into app files and exposed credentials, skip `.env` files and source maps, and an extra AI review can block more serious issues. You can share the finished link with specific people, a group or your whole team, all behind Cursor sign-in; on individual plans you can also share an app publicly. Team admins can see every app and secret, add their own review rules and unpublish anything. It works from a new Cursor Origin repo or an existing Origin or GitHub repo (GitLab is planned). Local agents cannot do this — it has to be a cloud agent.

Why it matters: most vibe-coded tools die on localhost. This puts hosting, a database and sign-in in the same step as writing the app, which is the difference between a demo on your laptop and something a teammate can actually open.

TRY IT → available now on paid plans that include cloud agents. On The X Subscription (Cursor's current plans) that means Premium at US$30 a month (about A$43.62) and above; the US$8 Plus plan (about A$11.63) does not include cloud agents. If you are still on a legacy Cursor Pro plan, that is US$20 a month (about A$29.08) and is the comparable tier. Start a cloud agent and tell it you want to build an app. Cursor: Build Live Apps · Cursor Apps docs · Cursor pricing

Dub Party, from Blazeej and Gamma Jack and published by Kestrel Interactive, is a party game about replacing every voice in a video with your own. You pick a built-in workshop scene or import a clip, claim the roles, record one line at a time in the booth, then watch the premiere together. You can play solo, pass one mic around on the couch (up to eight people, with friends claiming roles from their phones via a QR code and no extra app), or go online with up to 16 players over Steam. Optional Auto Dub runs on your own PC: it splits out the dialogue, transcribes it and suggests who is speaking, and the developers say importing a scene never uploads your video to an outside AI service. It unlocked on Saturday.

Why it matters: this is a low-stakes way to mess around with friends, and the on-device Auto Dub is a practical use of AI that does not send your clips to someone else's servers.

TRY IT → out now on PC (Steam) for A$5.56, down from A$6.95 (20% off at check time). Windows and Linux. Single-player, couch play and online. No Australian classification was listed at check time. Steam: Dub Party

A day after Live Apps, Cursor shipped Voice Chat with Agents: a realtime two-way call, not just dictation. You describe a change out loud, ask what an agent is working on, and hear short spoken updates while it works. It runs in the Agents Window, the IDE, on the web and in Cursor for iOS, where you can also say "Hey Siri, call Cursor" and mute or hang up from the Lock Screen. Under the hood the voice side runs on Grok Voice (voices include Ara, Eve, Leo, Rex and Sal); the coding agent in the chat keeps using whichever model you already picked for it. With Privacy Mode on, Cursor says voice chat runs with zero data retention. This is separate from the older Voice Mode, which only turned speech into text in the chat box. You need to be signed in — sessions that use only an API key cannot start a call — and there is a daily limit on voice calls.

Why it matters: checking on a long agent run, or briefing the next change, is often faster out loud than typing, especially on your phone while the agent keeps working.

TRY IT → update Cursor, sign in and click the waveform button labelled Start voice chat (it appears when the agent is idle and the chat input is empty). Cursor's docs require a signed-in account and mention a daily call limit, but they do not list a separate fee or name a paid-only plan. Cursor: Voice Chat with Agents · Cursor docs: Voice chat

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].