Meta's Connect developer pitch: Model API goes GA, open-weights Glimmer 30B, and a $1M hackathon
Meta's Connect developer sessions made the Model API generally available globally, shipped a 30B open-weights model that runs on a Mac Mini, took Muse Code out of beta, and opened a $1M hackathon.
After two days of hardware headlines at Meta Connect 2026, the developer sessions on September 24 turned the story into a build roadmap. Meta's recap reads as a single argument: frontier intelligence is only useful to developers at the speed it becomes tooling — so here's the tooling, most of it usable now.
The centerpiece is the Meta Model API, now generally available globally — Meta's first-party API plus third-party partners, with the enterprise support and compliance posture that the preview lacked. Alongside it, Muse Spark 1.3 is now available through the Oracle Cloud AI Platform, with Google Cloud in private preview.
For coding work specifically, three things landed. Muse Code is out of beta and available on Windows, positioned for long-horizon engineering: it's multi-agent by default, with parallel workers writing and testing code in isolated workspaces. Muse Glimmer is the more interesting release for tinkerers — an open-weights, 30B-parameter model designed to run on a single GPU or a Mac Mini, aimed at developers who want local control or on-device agents. And the Muse Connector Platform lets developers put their service inside Meta AI conversations via an API or MCP server, no app required — the same connector then travels with Meta AI onto new surfaces.
To get developers building immediately, Meta announced a Global AI Developer Hackathon: a ten-day virtual event with free Model API credits, hands-on workshops run by Meta Superintelligence Labs engineers, separate tracks for indie developers and startup teams, and a $1M cash prize pool. Applications open at dev.meta.ai.
The glasses and VR tracks got their own developer surface area. The Wearables Device Access Toolkit hit 1.0 after a year in developer preview, letting mobile developers extend existing apps to AI glasses without rebuilding — one integration covers display and displayless glasses, with camera capture, voice streaming, and motion tracking. Web developers get a lower-friction path: standard web APIs now render on Meta Ray-Ban Display with no native app or SDK, and WebMCP in developer preview lets Meta AI complete tasks inside a web app using only the functions the developer exposes. For the first time, Meta says, things built for glasses will be discoverable on-device and in the Meta AI app.
On VR, the v207 SDKs for Unity, Unreal, Android, and Web are available now so developers can build and test for the Meta VR Glasses (hardware ships spring 2027) before it exists — including gaze-and-hands input with no controllers in the box. Notably, the VR CLI now gives coding agents platform documentation, device debugging, and profiling, and a new Meta XR Operator lets an agent see, navigate, and interact with a running app in the simulator to fix bugs or build features and verify with a screenshot. Agents aren't just a product category at Meta anymore; they're part of the dev toolchain.
And for game builders, Horizon Create (mobile) and Horizon Studio (web) turn a described idea into a playable game using the Meta Horizon Engine's agentic creation — no game engine required — publishable natively across Facebook, Instagram, and Horizon feeds. A separate 2026 Meta VR Start Developer Competition runs September 24 through November 18 with $1M across 20 cash prizes.
Meta's framing — "the next year belongs to the developers who start now" — is the usual platform marketing, but the substance here is unusually concrete: a GA API, an open-weights model that fits on commodity hardware, agent-native VR tooling, and real prize money. The pieces that matter for working developers are the ones available today, and Meta was at least honest about which those are.