Meta has released Muse Spark 1.1, a significant update to its artificial intelligence model and a direct push into the competitive AI coding market. The company is positioning the new release against established offerings from OpenAI and Anthropic, with Meta Superintelligence Labs chief Alexandr Wang calling the pricing “very aggressive and attractive” relative to comparable products.
What Meta is launching
Muse Spark 1.1 is described by Meta as its strongest model for agentic and coding tasks to date. The first version, released in April, was limited to a small group of partners using a private API preview. The updated model is now being made available through a Meta developer portal as part of a public preview, where developers can sign up, review integration instructions, and join a waitlist for access.
For now, Meta is keeping API access on its own properties rather than distributing the model on third-party platforms such as OpenRouter. A company spokesperson said some early partners already have access and that new users will be added from the waitlist over time.
Pricing and developer access
Every new API account will begin with $20 in free credits, according to Wang. Beyond those credits, Meta will charge $1.25 per million input tokens and $4.25 per million output tokens. Wang framed the structure as a way to keep prices attractive as usage scales up.
Muse Spark 1.1 was trained to perform well on tasks involving third-party coding products and services, and Wang said the model is designed to work with the most popular developer harnesses in use today. Meta trained the model with coding capabilities in part because those skills improve overall agentic performance, where AI agents can carry out sequences of tasks with limited human supervision.
Why coding tools matter right now
The push into coding agents coincides with renewed interest across the tech industry in AI agents that can autonomously handle multi-step work. The article references OpenClaw, a tool that gained traction earlier in 2026 for managing the models behind more capable digital assistants.
From open-weight releases to paid APIs
Meta’s earlier AI strategy leaned heavily on releasing its Llama family of models to the open-source community. The Muse Spark rollout reflects a different approach: selling access to proprietary models through a paid API. Wang said Meta remains committed to open source and that an open-weight variant of Muse Spark is in development within MSL, though he declined to share a release date.
Broader context for the launch
The Muse Spark update comes the same week Meta introduced Muse Image (previously code-named Mango), an image generation model aimed at creators and advertisers. Meta is also training a larger model code-named Watermelon, though no release window has been announced. The Muse Spark model itself had been internally referred to as Avocado.
Wang, who leads Meta Superintelligence Labs, said he has been personally testing Muse Spark 1.1 on tasks such as searching the web, reading academic papers, and accessing personal health data, framing those workflows as a strong fit for agentic systems.
For developers watching the space, the practical takeaway is that Meta now has a publicly previewable API aimed at coding and agentic workloads, with introductory credits, published token pricing, and the promise of an open-weight variant to follow. The longer-term question is whether the pricing and developer experience will be enough to shift usage away from incumbent offerings from OpenAI and Anthropic.
Run a free scan to see your AI Visibility Score, SEO rating, and local citation accuracy.