Meta Platforms Inc. has launched Muse Spark 1.3, its most advanced artificial intelligence model to date.

  • Muse Spark 1.3 offers greater efficiency, improved agentic capabilities, and stronger coding skills than top industry competitors. 
  • Available to developers via the Meta Model API at the same price as its predecessor, the paid model will soon integrate into Meta's flagship consumer products, including Instagram, Facebook, and Meta AI. 
  • Meta is holding off on deciding whether to publicly release the model's weights while continuing development on its next major AI project, "Watermelon.” 

Meta Platforms Inc. (META) on Wednesday introduced its newest flagship artificial intelligence model, Muse Spark 1.3, to rival market leaders OpenAI and Anthropic.

The updated model, made available to developers on Wednesday through the Meta Model API, represents the company's "biggest jump so far on model performance," according to Alexandr Wang, Meta’s chief AI officer. 

The technology will soon roll out across Meta's family of social platforms, including Facebook, Instagram, and the Meta AI assistant. Wang said Muse Spark 1.3 demonstrates significant improvements in writing code and executing agentic tasks, in which AI independently performs multi-step actions on behalf of users. 

According to Wang, the model performs competitively against Anthropic's Claude Fable 5.1 and surpasses OpenAI's current GPT-5.6 Sol model in coding tasks.

META stock gained about 2.5% on Wednesday to record its best single-day jump in nearly a month. 

Business Strategy And Pricing

The release comes as Meta faces increased investor scrutiny over its massive AI infrastructure investments. Under Chief Executive Officer Mark Zuckerberg, the Menlo Park-based company restructured its internal AI division into Meta Superintelligence Labs (MSL) and appointed Wang to lead the unit last year.

Muse Spark 1.3 will be offered at the same price tier as Muse Spark 1.2. According to Wang, developer adoption for Meta's API services has grown rapidly, with some users processing trillions of tokens weekly.

While Meta has historically championed open-source AI development, Wang said the company has not yet decided whether to publicly release the model's open weights, its internal blueprint. However, Meta still intends to publish the weights for Muse Spark 1.2.

Enhanced Efficiency And Rigorous Safety

Muse Spark 1.3 offers several architectural improvements over earlier versions. The model requires 25% fewer tokens to perform operations, improves context retention over lengthy instruction sets, and supports concurrent workflows within a single session.

Safety mechanisms have also been bolstered. Meta conducted extensive safety evaluations after a prior incident in which an earlier model unexpectedly breached an external system during internal cybersecurity testing. The new model features heightened self-awareness of its limitations and asks for human confirmation before executing potentially permanent actions.

Looking ahead, Meta is continuing development on a larger, highly anticipated next-generation model internally code-named "Watermelon". While Wang declined to provide a release timeline, he said he is confident the upcoming model will be highly competitive in the rapidly evolving AI landscape.

META Stock: Retail View

Retail sentiment on Stocktwits was ‘bearish’ with ‘normal’ message volumes. 

META stock has lost nearly 10% year-to-date. 

For updates and corrections, email newsroom[at]stocktwits[dot]com.<