Stock Markets April 8, 2026 12:18 PM

Meta’s Muse Spark Spurs 7% Jump in Shares as Company Unveils New AI Model

Muse Spark debuts with multimodal reasoning, physician-curated health training data and a private API preview amid broader tech strength

By Nina Shah META
Meta’s Muse Spark Spurs 7% Jump in Shares as Company Unveils New AI Model
META

Meta Platforms shares rose 7% on Wednesday after the company launched Muse Spark, the first model in its Muse family developed by Meta Superintelligence Labs. The release follows a multibillion-dollar reorganization of Meta’s AI operations. Muse Spark offers native multimodal reasoning, tool-use support, visual chain of thought and multi-agent orchestration, and is available via meta.ai and the Meta AI app with a private API preview for select users.

Key Points

  • Meta launched Muse Spark, its first public AI model in the Muse family from Meta Superintelligence Labs; shares rose 7% amid broader tech strength.
  • Muse Spark offers multimodal reasoning with tool-use support, visual chain of thought, multi-agent orchestration, and is available at meta.ai and the Meta AI app with a private API preview for select users.
  • Meta rebuilt its pretraining stack over nine months, claiming similar capabilities to Llama 4 Maverick while using over an order of magnitude less compute; the model includes health reasoning trained with contributions from over 1,000 physicians.

Market reaction

Meta Platforms shares climbed 7% on Wednesday following the public introduction of Muse Spark, the company’s newest artificial intelligence model. The stock move occurred alongside broad strength across technology equities and coincided with the debut of the model, which Meta identifies as the initial offering in its Muse family created by Meta Superintelligence Labs.


Model capabilities and availability

Muse Spark is designed with native multimodal reasoning capabilities. That includes support for tool use, a visual chain of thought, and multi-agent orchestration. The company has made the model accessible via meta.ai and through the Meta AI app, and it has opened a private API preview to a select group of users.


Contemplating mode and benchmark results

Alongside Muse Spark, Meta introduced a feature called Contemplating mode, which runs multiple agents that reason at the same time. In benchmark testing cited by the company, the mode produced a 58% score in Humanity’s Last Exam and a 38% score in FrontierScience Research.


Engineering and efficiency

Meta said it rebuilt its pretraining stack over a nine-month period, addressing model architecture, optimization routines, and data curation. The company reports it reached the same capabilities as its previous model, Llama 4 Maverick, while using over an order of magnitude less compute.


Applications and health work

The company highlighted multimodal visual integration across domains and health reasoning among Muse Spark’s applications. For health-related outputs, Meta said it collaborated with more than 1,000 physicians to help develop training data. The model is capable of producing interactive displays that explain nutritional content and illustrate which muscles are engaged during exercise.


Safety testing and third-party evaluation

Under Meta’s Advanced AI Scaling Framework, Muse Spark was evaluated for safety in high-risk areas and demonstrated strong refusal behavior for content related to biological and chemical weapons. Third-party testing by Apollo Research reportedly found Muse Spark exhibited the highest rate of evaluation awareness among models they have examined.


Note: The information above reflects details provided by the company regarding the model, availability, engineering work, benchmarks, health collaborations, and safety assessments.

Risks

  • Availability is initially limited - the private API preview is open only to select users, which may constrain broader adoption and enterprise integration in the near term.
  • Safety remains a focus - while Muse Spark demonstrated strong refusal behavior in high-risk domains under Meta’s testing framework, high-risk content such as biological and chemical weapons necessitates ongoing evaluation.
  • Performance metrics are reported on specific benchmarks (Humanity’s Last Exam and FrontierScience Research) and may not capture all use-case performance or real-world deployment limitations.

More from Stock Markets

Trio Petroleum Shares Plummet After Fresh At-the-Market Share Sales Apr 8, 2026 Plus Therapeutics Shares Jump After FDA Grants Orphan Drug Status to REYOBIQ Apr 8, 2026 U Power Shares Rally After $3.19M Share Subscription Agreement Apr 8, 2026 Compass Lexecon Sues Bill Hwang for $14.2 Million in Unpaid Defense Fees Apr 8, 2026 Lufthansa Cabin Crew Union Announces One-Day Strike Impacting Frankfurt and Munich Flights Apr 8, 2026