Credit: Getty Images |NurPhoto

Meta on Wednesday announced Spark, the first AI model in the Muse family that it says represents “a ground-up overhaul of our AI efforts.”

Muse Spark is the first release of Meta’s Superintelligence Labs, formed a little less than a year ago with the grandiose goal of “deliver[ing] on the promise of personal superintelligence for everyone.” The release represents a clean break from Meta’s previous work on the open source Llama model family, which has received a middling reaction both from users and on independent LLM rankings. And while Spark will be a proprietary model, Meta founder and CEO Mark Zuckerberg said in a post on Threads that the Muse family will “includ[e] new open source models” in the future.

Meta said that Muse Spark will take advantage of content posted across platforms like Instagram, Facebook, and Threads, much as xAI’s Grok is integrated with content posted on X. Currently, this means Muse Spark can link to public posts related to a location or trending topic that you ask about, for instance. In the future, Meta says this will expand to “new features that cite recommendations and content people share” and “Reels, photos, and posts woven directly into your answers, with credit back to the content creators.”

Contemplation and compression

Meta says Muse Spark’s contemplation mode “enables superior performance with comparable latency.”
Credit: Meta

In a somewhat technical blog post accompanying the Spark announcement, Meta includes the by now routine laundry list of AI benchmarks, in which Muse Spark’s standard thinking mode ranks comparable or better than competing models from OpenAI, Anthropic, Google, and xAI. But that post also sheepishly acknowledges that “we continue to invest in areas with current performance gaps, such as long-horizon agentic systems and coding workflows.”

In that same post, Meta also touts “Contemplating” mode, which it says will be “rolling out gradually” and which can “orchestrate multiple agents that reason in parallel.” By using up to 16 agents thinking in concert simultaneously, Meta says that Contemplating mode “enables superior performance with comparable latency.” That “superior performance” includes a reported high water mark of 58.4 on Humanity’s Last Exam (with the use of external tools), according to Meta.

A Meta graph shows how additional training leads to “compression” of token usage before additional gains in accuracy.

And while previous Llama models faced criticism for not taking advantage of reinforcement learning, Meta says Muse Spark shows “smooth predictable gains” after additional RL steps after pretraining, “improving model reliability without compromising reasoning diversity.” That reinforcement learning system also makes use of “thinking time penalties,” which Meta says balance the need to “maximize correctness” with optimizing the number of tokens used. In testing on the AIME 2025 benchmark, Meta says it saw a “phase transition” where the model started compressing equally accurate reasoning into “significantly fewer tokens.” After that compression, subsequent trained models slowly increased the token usage again to achieve even higher accuracy in less overall time than the previous uncompressed versions.

The release of Muse Spark comes alongside an update to Meta’s Advanced AI Scaling Framework, which the company says now covers a broader range of potential AI risks. The company says that the model “falls within safe margins across all frontier risk categories we measured,” but says that more details will only be available in an upcoming Safety & Preparedness Report.

Muse Spark is available now in the Meta AI app and via the meta.ai website, as well as a private preview API for “select partners.” Meta says the model will be available via WhatsApp, Instagram, Facebook, Messenger, and AI glasses “in the coming weeks.”