Meta expands Muse Spark AI model with public API and developer pricing

By NewSunSEO | July 9, 2026

Meta rolled out Muse Spark 1.1, its strongest AI model yet for coding and agentic workloads, alongside developer pricing that chief AI officer Alexandr Wang called “very aggressive and attractive” against Anthropic and OpenAI. The updated model is now offered as a public preview through a developer portal, with early partners already granted API access and new users able to join a waitlist.

What Meta announced

Muse Spark 1.1, introduced on Thursday, is Meta’s strongest model yet for coding and agentic workloads, according to Alexandr Wang, Meta’s chief AI officer. The original Muse Spark was released in April and limited to select partners through a private API preview. The updated version is now available through a developer portal as a public preview, where users can sign up and review integration instructions. A Meta spokesperson said some early partners already have API access and that new users can join a waitlist to be onboarded over time.

For the moment, Meta is restricting API distribution to its own properties rather than listing the model on third-party marketplaces such as OpenRouter. This is going to be served on top of the computer infrastructure that we’ve built, Wang told CNBC.

How is pricing structured for developers?

Wang described the pricing as very aggressive and attractive relative to comparable offerings from Anthropic and OpenAI. Each new API account begins with $20 in free credits, with input charged at $1.25 per million tokens and output at $4.25 per million tokens. The aim, Wang said, is to have attractive pricing that scales with immense consumption usage.

He added that Muse Spark 1.1 outperformed rival models on certain tasks involving interaction with third-party coding products and services, though he did not name specific benchmarks.

Why coding matters for agentic AI

Muse Spark 1.1 was trained inside Meta Superintelligence Labs, or MSL, with an emphasis on coding because strong code generation underpins the broader agentic capabilities Meta is pursuing. You kind of have to build coding capabilities as part of that in service of overall agentic capabilities, Wang said, describing AI agents that could eventually perform multiple tasks like a fleet of human interns.

Wang said Meta designed the model to work well with the popular developer harnesses in use today, calling that the best approach for this model given our goal to maximize adoption. The reference comes amid excitement across the tech industry around AI agents, fueled in part by the popularity of OpenClaw for orchestrating AI-powered digital assistants.

How do open source and proprietary models fit together at Meta?

Although Meta historically emphasized open-weight releases through its Llama family, the company is now selling access to proprietary models as well. Wang said MSL still intends to open-source a variant of Muse Spark that is currently in development, though he declined to share a release timeline. He also confirmed that Meta is training a more powerful model, code-named Watermelon, without offering a launch date. Muse Spark itself was previously known internally as “Avocado.”

Broader context at Meta

The rollout is part of a wider effort to diversify Meta’s business beyond advertising. The company is under pressure from Wall Street to demonstrate a return on its substantial and growing investment in AI infrastructure and development, particularly since it lacks a cloud infrastructure business comparable to those operated by its largest competitors. Earlier in the week, Meta also released Muse Image, a text-to-image model originally code-named Mango, aimed at creators and advertisers.

Wang said he has been personally testing the latest Muse Spark for health-related tasks, such as searching the web, reading academic papers, and accessing personal health data, and views that as a strong early use case for agentic systems.

FAQ

What is Muse Spark 1.1 and who leads the effort?

Muse Spark 1.1 is Meta’s AI model focused on coding and agentic workloads, released on Thursday as a public preview through a developer portal. It was developed inside Meta Superintelligence Labs and described by chief AI officer Alexandr Wang as the company’s strongest model yet for those tasks.

How much does the Muse Spark API cost?

Each new API account starts with $20 in free credits. Input is priced at $1.25 per million tokens and output at $4.25 per million tokens, a level Wang characterized as “very aggressive and attractive” compared with Anthropic and OpenAI.

Will Meta still release Muse Spark as open source?

Yes. Wang said Meta Superintelligence Labs intends to open-source a variant of Muse Spark that is currently in development, though he declined to share a release timeline.