Meta’s Superintelligence Lab launched a pair of new state-of-the-art models on Monday, Muse Marvel and Muse Marvel Full, representing the corporate’s first try at a frontier mannequin.
The place Muse Spark aimed to be a light-weight mannequin that would carry out at low latency, the Marvel fashions are bigger and mentioned to be extra succesful, with increased benchmark scores and higher outcomes on complicated duties like coding.
Based on Meta’s testing (introduced in the model card), Marvel and Marvel Full obtain comparable benchmark scores to Fable 5.1 and GPT-6 Astra.
“Collectively, they mark the following step on our scaling ladder and a significant leap towards private superintelligence,” Meta’s weblog submit concerning the fashions reads.
That scale is mirrored within the pricing: Accessed by way of the Meta Mannequin API, Marvel will value $4 per million enter tokens and $20 per million output tokens, making it roughly equal in value to GPT-6 Sol or Opus 5.5.
Notably, Marvel Full will solely be accessible to Muse Code subscribers and won’t be served by way of the API. Each fashions may even be used to energy the Muse agent.
In step with Meta’s curiosity in creating private assistants that may carry out duties over a protracted interval, the corporate positioned specific emphasis on the brand new fashions’ means to retain and synthesize info given by the consumer, a property Meta calls “energetic reminiscence.”
To attain higher recall, Meta highlighted the MemoryBench benchmark, which challenges continuous studying and reminiscence development frameworks. Based on the corporate’s testing, Muse Marvel now scores increased on MemoryBench than Fable, Opus, or Astra.
“We put an amazing quantity of effort into ensuring your mannequin can bear in mind what’s vital to you over days, weeks, months, and past,” mentioned analysis scientist Kavya Kopparapu in a video accompanying the launch.
The brand new fashions are the primary frontier mannequin launch since Scale AI founder Alexandr Wang restructured Meta’s AI efforts in August of last year. Since then, Meta has launched its light-weight Muse Spark mannequin and Muse agent however has prevented the type of state-of-the-art mannequin releases OpenAI and Anthropic are recognized for.
Based on the submit, the fashions have been skilled at Meta’s Prometheus data center in central Ohio, which got here on-line earlier this 12 months.
Whenever you buy by way of hyperlinks in our articles, we may earn a small commission. This doesn’t have an effect on our editorial independence.
