Muse Spark 1.3’s public xhigh variant and its limited-preview Max variant are not interchangeable in a benchmark headline.

Watch the selected video on YouTube

Meta’s Muse Spark 1.3 is a useful reminder that a launch headline can flatten important product distinctions. Artificial Analysis’ September 2 write-up gives the publicly available xhigh variant an Intelligence Index score of 61. It gives the Max variant a score of 62, but says Max is in limited preview for Meta partners. VentureBeat highlights the same practical caveat: the best reported results are not necessarily available to ordinary developers on the same terms as the public model.

That does not make the public release uninteresting. Artificial Analysis reports gains in agentic-work and scientific evaluations, a one-million-token context window, and the stated xhigh pricing. But a fair comparison must keep variant, benchmark and availability aligned. “Muse Spark 1.3 scored 62” is incomplete if the reader is evaluating the model they can actually call today.

The optional ByteForward video is included for release-day context only. The review does not use it as proof of benchmarking, cost or public availability. The substantive evidence comes from the written analysis and the supporting reporting.

Human-review risk. Benchmark scores and cost figures are attributed to Artificial Analysis; the draft does not treat them as universal real-world performance claims or extrapolate Max’s limited-preview score to all developers.

Bottom Line

Muse Spark 1.3’s public xhigh variant and its limited-preview Max variant are not interchangeable in a benchmark headline.

Sources