Meta Muse Wonder launches, but the API model is not the flagship

Meta has earned a place in our evaluation queue, not in our default production route. Muse Wonder matters because it turns Meta into a credible frontier API supplier, giving software teams more leverage against concentration and vendor lock-in. But the launch is commercially awkward: the model developers can actually call costs $4 per million input tokens and $20 per million output tokens, while the stronger Muse Wonder Full is confined to Muse Code subscriptions. Independent Artificial Analysis results put Wonder at 52 and Full at 56 on its Intelligence Index, so the capability Meta most wants us to notice is not the capability we can deploy behind our own product. At Virtual Arc, we would add Wonder behind an existing routing interface, shadow-run long-horizon coding and document-heavy agent tasks, and measure cost per accepted result, retries, p95 latency and human intervention. We would not rewrite an agent stack or move core traffic until the callable model shows a repeatable workload-level win. A new frontier vendor is strategically important; this specific release is still a benchmark candidate, not a migration event.