The release season is still going strong: following yesterday, the rumored Gemini 3.8 Flash has also arrived today. However, Muse Spark 1.3—the model promised in Zuckerberg’s big comeback open letter last month—is absolutely headline-worthy today. According to data from AAII, it is already ranked the No. 3 model in the world (?!?!).
Take a look at the results they’ve finally achieved—performance that can rival the frontier models from OpenAI and Anthropic (it’s Opus, not Fable)—as well as the confidence they’re showing… Even more excitingly, they’ve promised that Muse Spark will be released with open weights (!!!):
Muse Spark 1.3 begins rolling out today, with frontier-level performance at nearly negligible cost. This is by far our biggest leap yet in coding and agentic work. Try it in Muse Code and through our API.
Next up is 🍉, with the open-weight version of Muse Spark coming soon.
They’ve adopted a very interesting pricing model: if you opt to allow the model to be used for training, the price is reduced by more than 90%:
