Friday roundup: A good week for coding, speech models

OpenAI makes Realtime API generally available
The agentic Realtime model is a native speech-to-speech model that can be used to make customer service agents, phone reps and voice navigation features. It doesn’t go through speech-to-text and text-to-speech loops and generates audio «directly through a single model and API.» OpenAI is marketing this to developers who want more natural flowing speech, and it’s not available as distinct model in ChatGPT – yet. You can hear it and see it in use at places like Zillow, T-mobile, StubHub and Oscar Health, though. With general availability, it will surely show up in a lot more places soon.
More at: OpenAI’s launch page, discussion on r/OpenAI.

Read on for more news!

xAI launches Grok for coding — and it’s fast
The Grok 4 model launch didn’t include the whole feature set, opting instead for a staggered release of sub-models with specific purposes. Today’s agentic coding model — the grok-code-fast-1 — is built from scratch on their 200,000 GPU «supercomputer» and is meant to massively speed up repetitive day-to-day coding tasks. It’s available for free «for a limited time» on Cursor, GitHub Copilot, Windsurf and whole host of IDEs. It is also dirt cheap when it starts charging.
More at: xAI’s launch page and a writeup by Reuters.

Microsoft unveils first homegrown models
The company believes that voice communication is the future of AI, and is releasing MAI-Voice-1, a «highly expressive and natural speech generation model,» that can run on a single GPU and is available in Copilot Daily and Podcasts, and a new Copilot Labs Audio Expression. The other, text-based model is the MAI-1-preview, which is the first model trained by Microsoft from end-to-end. It’s available on LMArena and is being rolled out slowly.
More at: Microsoft’s launch page, writeups at Engadget and The Verge.

Apple’s Xcode now supports Claude
Xcode is updating to support the latest operating systems from Apple, but underneath the hood you will find and option to integrate your Claude account and use Claude Sonnet 4 as your assistant. It also now supports GPT-5, and GPT-5 (Reasoning) from the OpenAI API. The final version will support both models, and it should be possible to bring API keys from other providers.
More at: Apple’s release notes, MacRumors.

Lots of upgrades for OpenAI’s Codex, by popular demand
The agentic coding app is getting a GPT-5 facelift this week, with other goodies. It now has new IDE extension, supporting VS Code, Cursor and other forks directly in the environment. You can also seamlessly upload your local work to the cloud and the other way around, to keep working in the best environment for any task. Codex also gets a new UI that supports images, message queuing, to-do-lists and web searches, to name a few improvements.
More at: OpenAI’s launch thread.