Several points remain open. Meta's benchmarks compare the model mainly with US rivals such as Thinking Machines' Inkling and Nvidia's Nemotron 3 Ultra, and with Chinese open models. Those results come from Meta, and independent evaluations have not yet been published. Reporting says the weights sat in a gated repository behind an access-controlled form for four weeks before the public release. The reason for that period is not explained. The license is also not fully open. Its full terms beyond the requirements noted below are not yet analyzed in public reporting.
For attorneys, the license is the main item to review. It requires a visible notice in every product that uses the model and a separate agreement for large model-as-a-service providers. Teams evaluating the model for commercial use should confirm which of those obligations apply to them before treating "open" as a permissive license. Procurement and vendor contracts should also account for the cloud catalog listings and hosted API access, since those channels may carry different terms. The hardware requirements make local deployment a capital decision, not a casual one. Watch for independent benchmark results and any clarification of the license scope from Meta.