[AINews] Muse Spark 1.3 matches GPT-5.6-Sol, confirming Meta Superintelligence as the newest Frontier Lab, >90% discount for training
Latent Space ● Covered by 7 sources
Muse Spark 1.3 just jumped into the frontier-model chat, and Meta says it’ll ship open weights soon. The odd twist: it’s 90%+ cheaper if you let it train on your data.
Based on reporting by Latent Space — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Meta’s Muse Spark 1.3 is the loudest release in this batch. The model is being framed as the strongest Spark variant for agentic and coding work, and the chatter around it is unusually blunt: the public benchmark screenshot puts it alongside models like GPT 5.6 Sol and Opus 5, while AAII now lists it as the world’s #3 model. That is a serious claim, not marketing fluff tucked into a blog footer.
The part that changes the story is openness. Muse Spark is promised as open weights, which is exactly the sort of line that makes the closed-model crowd sit up and check the benchmarks twice. Meta also appears to be betting that pricing can be a weapon here: the model is reportedly more than 90% cheaper if customers opt in to training. That is not a tiny discount. That is a dare.
This release fits the bigger pattern from the past few days: the frontier is getting noisier, and the lines between “best” and “available” keep shifting. Open model teams are no longer just chasing niche wins or hobbyist goodwill. They are showing up with real numbers, real pricing pressure, and enough confidence to compare themselves against the established heavyweights.
What matters most is not just that Muse Spark 1.3 looks competitive. It’s that Meta is now acting like a Frontier Lab in the same way the old guard does — with model launches, price hooks, and a promise of open weights all in one move. That’s a different posture from the usual “we built something impressive, please clap” routine.
My take — AI-written commentary, not fact-checked reporting
Open weights are the right bet here, because closed labs have had long enough to charge luxury-car prices for software that still sometimes acts like a confused intern. The 90%+ training discount is the real headline; it turns model access into a business decision instead of a loyalty test. That’s how you pry a market open.
Read more about this at: Latent Space