G'day, and welcome to the first issue. Five releases worth your time, why each one matters, and how to try it. About a five-minute read.

xAI's new speech-to-speech model is live in its Voice Agent API at US$0.08 a minute (about A$0.11, or roughly A$6.90 an hour), the same price as Grok Voice 2.0. Apps using the grok-voice-latest alias have already been switched over, and Grok Voice 2.0 is now deprecated. xAI cites Artificial Analysis testing that puts it #1 on the Speech-to-Speech Index (86.9), with responses in under a second.

Why it matters: if you're building a voice agent or phone bot, you get a better model at no extra cost and with no code change.

TRY IT → set the model to grok-voice-3.0 (or grok-voice-latest) in the xAI API. x.ai: Introducing Grok Voice 3.0 · xAI release notes

StepFun has published the open weights for Step 5 Preview, its flagship model for agentic work: 600B parameters (27B active), up to 1M tokens of context, and text and image input. It's under the StepFun Model License 1.0, which is Apache-style with extra attribution and commercial-licence conditions.

Why it matters: open weights at near-frontier scale, pitched at coding and finance agents. Be realistic, though: the smallest listed setup is one server with 4× B300 or 8× H200 GPUs, so it's for teams and cloud renters, not laptops. It's also available through StepFun's own apps and API.

TRY IT → Hugging Face: stepfun-ai/Step-5-Preview (FP8; BF16 and NVFP4 versions are also up) · StepFun model page

At Config India in Bengaluru, Figma launched Specs, a shared Markdown document that lives on the canvas next to your designs. You write down requirements, decisions and design rules once, and Figma's agent and outside coding agents (through Figma's MCP server) can read and update it. Code layers, which put live, code-backed prototypes on the canvas, are now in open beta, and you can make one from a design frame, a GitHub repo, your website or a prompt. Figma's MCP server also added support for 20 new partner tools.

Why it matters: vibe-coded prototypes have a habit of ignoring your design system. This puts the rules where the agents can see them, inside the design tool.

TRY IT → Specs is rolling out in open beta on paid plans. Code layers is rolling out to everyone over the next few weeks, and creating and editing code layers is free on paid plans until 4 November, then uses AI credits. Figma: Introducing Specs · Figma: everything from Config India

Epic has released a preview of its rebuilt Games Launcher for Windows, macOS and, for the first time, Linux. Epic says it's faster and lighter, and its own announcement says the team "vibe coded" it. An Epic executive told The Verge that agents wrote and reviewed all the code, with developers supervising. The old launcher stays available, and Epic says Fortnite will come to Linux through the launcher once its anti-cheat rebuild is ready.

Why it matters: this is one of the biggest consumer apps a company has openly said was built by AI agents, and Linux gamers finally get a native Epic client.

TRY IT → free download, sign in with your Epic account. Epic Games Store: the new launcher in Preview · background from The Verge

Google Research has opened Deep Review to the public. It's an agentic reviewer that splits a long document into sections, checks each one in parallel for logic gaps and unsupported claims, then pulls it all together into one report. Everyone gets two free reviews a week (more if you connect a Gemini API key), and it takes PDFs and plain text for now. Google says uploaded documents are never used to train its models and are kept for a year so you can see them in the app.

Why it matters: a free, rigorous second read of a spec, proposal or contract before it goes out.

TRY IT → free, two reviews a week. Google Research: Deep Review

Also shipped

Skip this: "#1 on the leaderboard" launch-day claims. This week's vendors all quote the rankings that suit them. Run one of your own real tasks before you switch.

Five to Try is curated with AI help and checked by a human. Hit reply and tell us what you tried.