Maya: Good morning, and welcome to Meta Developer News, for Monday, October fifth, twenty twenty-six. I'm Maya.
Devon: And I'm Devon. Here are the developer stories that matter across Meta's platforms today, starting with a major Llama release.
Maya: About two weeks ago, Meta released Llama three point two, a new family of open models built for the edge. There are lightweight text models at one billion and three billion parameters, and small to medium vision models at eleven billion and ninety billion parameters, all with a one hundred twenty eight thousand token context window.
Devon: That is a big deal for on-device A I. The one billion and three billion models are small enough to run inside a phone's memory, so you get real instruction following and summarization without a round trip to the cloud. And the eleven and ninety billion versions bring vision into the mix.
Maya: Exactly. Meta packaged the release with two developer tools that matter. First, the Llama Stack framework, which standardizes how you build with the models. And second, the PyTorch ExecuTorch runtime, which is the turnkey path to running these models on device.
Devon: And they lined up day-one hardware support, with optimizations from Qualcomm, MediaTek, and Arm. So if you are shipping a mobile or embedded A I feature, the toolchain is ready at launch, not six months later.
Maya: One more data point on Llama's reach. A Counterpoint report covering the first half of this year found that Llama base models power thirty eight percent of the adapted sovereign large language models surveyed. That makes Llama the most used foreign base model across regions. Open weights remain Meta's biggest developer moat.
Devon: Moving to X R, Horizon O S version two point nine is rolling out in the Public Test Channel, and it brings something developers have asked for: a real window manager. You can now place three Quest windows side by side, or stack them.
Maya: Three panels is not exactly a thirty-monitor battlestation, but it is a meaningful step. A browser, a store page, and a video can finally coexist. And windows now launch spatially, which Meta says makes them easier to drag and position.
Devon: The practical takeaway for developers: if your app assumes it always has the full display, now is a good time to test your layouts in multi-window mode. The features started rolling out the week of September twenty eighth, so Test Channel users should be seeing them now.
Maya: And one more tool update for Quest developers. The Meta Quest Developer Hub just got a significant refresh. The Performance Analyzer trace capture now uses named presets instead of a long list of options. There is a low-overhead C P U preset, a G P U surface and metrics preset, and a render pass diagnosis preset, so you pick your intent and go.
Devon: Two fixes in there that developers will feel immediately. G P U profiling is now a per-device opt-in instead of always on, so your first captures stay clean. And the metrics that showed up blank on recent Horizon O S versions, including on Quest three, now display real values again.
Maya: Plus casting reliability improvements and a unified app updater. If you have been fighting the Developer Hub lately, this update is worth a look.
Devon: That is it for today's briefing. If you are building with Llama, send us a link to what you are working on. We read everything.
Maya: We will be back tomorrow with more. I'm Maya.
Devon: And I'm Devon. Have a great build day.