Meta has launched a significant update to its Muse Spark large language model and revealed its pricing strategy.
The company's AI chief stated that the pricing is highly competitive and attractive compared to similar offerings from Anthropic and OpenAI.
This move introduces a new monetization channel for Meta's diversified business portfolio, as developers will need to pay to access this new AI model.
Key Developments in the AI Arena
In San Francisco on June 4, 2026, Meta Platforms, Inc. (META) Chief AI Officer Alexander Wang presented at a technology summit.
Three months after introducing its first self-developed large model under Wang's leadership, Meta has released a major version update, aiming to compete directly with OpenAI and Anthropic in a core industry segment.
Wang stated that Muse Spark 1.1, officially launched on Thursday, is currently Meta's most powerful model for autonomous intelligent agents and code generation.
The initial Muse Spark model in April was available only to a select group of partner companies via a private API preview.
Access and Infrastructure
The API for the new model is now available for public trial on a developer platform, where users can register and review deployment guides.
A Meta spokesperson indicated that some early partners already have access, while new users can join a waitlist for phased rollout.
The company noted that the interface is currently only for its own products and will not be available on third-party platforms like OpenRouter at this stage.
Wang emphasized that the model will be served using Meta's self-built computing infrastructure.
This marks the second major product in Meta's Muse series this week, following the Tuesday release of the Muse Image generation model, aimed at attracting creators and advertisers.
Market Pressure and Strategic Shift
Meta faces ongoing pressure from investors to demonstrate a return on its massive and growing investments in AI infrastructure.
While its compute investments rival major cloud providers, Meta has not yet established a mature cloud business and has lagged behind OpenAI, Anthropic, and Alphabet (GOOG) in launching hit models and AI applications.
Wang stated that the pricing for the updated Muse Spark is aggressive and compelling compared to rival products.
New API accounts will receive $20 in free credits, with pricing set at $1.25 per million tokens for input and $4.25 per million tokens for output.
The core objective of this pricing is to create a cost-effective solution for large-scale, high-consumption scenarios.
Model Capabilities and Future Roadmap
He claimed that Muse Spark 1.1 outperforms competitors in integrating with third-party code tools and handling automated interactive tasks.
Meta's Super Intelligence Lab, led by Wang, focused on enhancing the model's coding capabilities to upgrade AI agents that can autonomously handle complex, multi-step tasks.
Code generation ability is described as an essential foundational support for building a complete autonomous agent system.
The AI agent sector gained significant traction in early 2026, partly driven by the popularity of tools like OpenClaw.
Wang noted that Muse Spark 1.1 is compatible with mainstream development frameworks, a path chosen to maximize product adoption.
This represents a strategic shift for Meta, moving from a core focus on open-sourcing its Llama series models to offering paid access to closed-source commercial models.
Wang clarified that Meta has not abandoned open-source, with the lab developing a derivative version of Muse Spark planned for future open-source release, though no timeline was provided.
Personal Applications and Next-Gen Models
Wang shared his personal testing of the new Muse Spark and expressed optimism about its potential in personal health applications, such as retrieving information, reading academic papers, and accessing personal health data for management.
He stated that such scenarios perfectly illustrate the core value of an intelligent agent system.
Wang also revealed that Meta is training a more powerful large model, internally codenamed "Watermelon," with no release date announced, while the first Muse Spark was internally codenamed "Avocado."
Comments