Meta this week put out Glimmer, an open-weight model you can download and run on your own hardware. At the same time, the company kept Muse Spark — its more capable system — locked behind proprietary APIs. Mark Zuckerberg framed the move with a letter arguing that AI should belong to everyone rather than a few labs. The reality is more nuanced: Meta is open where it costs little and closed where the real competitive edge lives.
That two-tier pattern matters because it tells you what you can and can't rely on. Open weights mean you can inspect the model, fine-tune it, run it offline, and avoid per-call API fees or vendor lock-in. But if Glimmer is the deliberately weaker tier, teams building serious products may still find themselves pulled toward the paid, closed model for anything demanding. "Open" here is a distribution strategy, not a gift of the best technology.
For builders, the practical takeaway is to treat Glimmer as a real asset for the right jobs. Local deployment is a strong fit for privacy-sensitive data, edge use cases, cost-controlled batch processing, and rapid experimentation where you want full control. Benchmark it against your actual workload before assuming you need the premium tier — the gap between an open and a closed model is often smaller than marketing implies for narrow, well-defined tasks.
Just as important: read the license. "Open weights" is not the same as open source, and Meta's releases have historically carried usage restrictions and acceptable-use terms. Before you build a business on top of Glimmer, confirm commercial rights, redistribution rules, and any scale limits, so a licensing surprise doesn't undo your architecture later.
The broader lesson is to judge every "open AI" announcement by what's actually released, not the rhetoric around it. Ask which model is open, how it compares to the company's flagship, what the license permits, and where the vendor still keeps you dependent. That framing turns a headline into a decision you can act on.