A week where a 30B open-weights release was narrated first by the hardware platform, not by Meta itself.
Meta's 30B coding model Muse Glimmer arrived with open weights, and the how-to for running agentic workflows locally came from NVIDIA's developer blog. The developer community picked it up on the Hacker News front page.
At 30B the model fits on hardware people already own, making it realistic to keep an agent loop entirely inside a closed environment. That the platform vendor, not the publisher, framed the release also suggests the axis of value is drifting from whose weights these are toward where they run. The bundled articles also include unrelated infrastructure-policy threads, so this is not confirmed as a single trend.
Reported real-world throughput and quantized builds, licensing and benchmark detail from Meta itself, and whether teams actually adopt it for day-to-day coding.