On July 9, 2026, Meta introduced Muse Spark 1.1, an AI model built for coding and agentic tasks, per Meta Superintelligence Labs. The launch paired the model with a public preview of a new Meta Model API, pricing built to compete against Anthropic and OpenAI.
Quick Summary – TLDR:
- Muse Spark 1.1, per Meta Superintelligence Labs, is a multimodal reasoning model with major gains in coding, tool use, and computer use over the original Muse Spark.
- The new Meta Model API, per Meta, entered public preview, letting outside developers build with Muse Spark 1.1 for the first time.
- Meta chief AI officer Alexandr Wang called the pricing “very aggressive and attractive” against comparable offerings from Anthropic and OpenAI, per CNBC.
- New API accounts start with $20 in free credits, then pay $1.25 per million input tokens and $4.25 per million output tokens, according to Wang.
- Muse Spark 1.1 can actively manage a 1 million token context window across extended coding and agentic workflows, per Meta.
What Happened?
Three months after Meta’s first AI model launch under Wang, the company rolled out Muse Spark 1.1 as a major update aimed at coding and agentic performance. Wang described it in a CNBC interview as Meta’s “strongest model for agentic and coding work yet“.
The model is available now in “Thinking” mode in the Meta AI app and on meta.ai. For now, Meta is limiting API access to its own properties rather than listing Muse Spark 1.1 on third-party marketplaces, with early partners already using it and new developers joining a waitlist over time. That staged rollout mirrors the caution large labs have shown while scaling AI agents beyond a limited partner set.
The release comes the same week as a companion image-generation model, bringing Meta closer to its stated vision of personal superintelligence.
(1) Today we’re releasing Muse Spark 1.1 — a strong agentic and coding model at a very low price. It’s available through our new Meta Model API and in Meta AI.
— Mark Zuckerberg (@finkd) July 9, 2026
Muse Spark 1.1’s Coding and Agentic Gains
Coding performance improved substantially on real-world tasks involving large, complex codebases, with the model able to diagnose and fix complex bugs, implement new features in enterprise-grade systems, and execute large code migrations. As a main agent, Muse Spark 1.1 gathers context, builds a plan, and delegates execution across parallel subagents; as a subagent, it stays within its assigned job and knows when to escalate back.
Amjad Masad, CEO of Replit, calling it a complete agentic foundation, said:
Wang’s Meta Superintelligence Labs, or MSL, trained Muse Spark 1.1 to excel at coding specifically because that skill underpins broader agentic capability, letting AI agents autonomously handle multiple tasks “like a fleet of human interns,” he said. This positions coding less as a standalone feature and more as the load-bearing skill behind Meta’s wider agent push into AI coding ecosystem.
Pricing Aimed Squarely at Anthropic and OpenAI
Meta is charging developers to access Muse Spark 1.1 through the Meta Model API, a shift from the company’s earlier strategy of releasing models to the open-source community. Wang said MSL still has a Muse Spark variant in development that it intends to open-source, though he declined to say when.
That tension between the proprietary API and Meta’s open-source roots is the clearest sign yet of how far the company will bend its own playbook to compete for the same developers who default to OpenAI and Anthropic’s Claude for agentic coding work today.
What’s Next?
Meta has not set a date for opening the Meta Model API beyond its own properties or for the open-source Muse Spark variant Wang referenced. Developers on the current waitlist will be added “over time,” and the pricing Wang called aggressive will face its real test once usage scales past the free-credit tier and against Anthropic’s and OpenAI’s own coding-model pricing moves.
SQ Magazine’s Takeaway
Undercutting on price while still walling the API off to Meta’s own properties is a hedge, not a full commitment to the open developer market Anthropic and OpenAI already compete for. Meta is testing whether aggressive per-token pricing can pull developers away from established coding assistants before it commits to wider distribution through marketplaces like OpenRouter.
The more telling signal is Wang’s framing of coding as instrumental to agentic capability rather than a product in its own right. That reasoning explains why Meta trained Muse Spark 1.1 to work across popular third-party harnesses instead of building a closed coding environment, a bet that agent orchestration, not a standalone code editor, is where the category is actually heading.
