For nearly two years, if you wanted Meta’s best AI model you had to use it through Meta’s own apps. That changed on July 9, 2026.

Meta Superintelligence Labs released Muse Spark 1.1 alongside the public preview of the Meta Model API — the first time outside developers can access a Muse Spark model directly. Businesses and developers in the US can now build on the same model that powers Meta AI, without going through Meta’s consumer products.

This is a significant shift. When Meta launched the original Muse Spark in April, it was a statement: after years of open-sourcing Llama, Meta was keeping its best model proprietary. But keeping it locked inside Meta’s apps meant no enterprise developer could actually integrate it into their own products. The Meta Model API fixes that.

What Muse Spark 1.1 Actually Is

Muse Spark 1.1 is a multimodal reasoning model built explicitly for agentic tasks. The context window is 1 million tokens — enough to feed it an entire codebase, a full legal contract set, or months of operational logs in a single call.

The model can process images, video, and documents natively. It handles multi-agent workflows both as a primary orchestrating agent and as a subagent taking instructions from another system. This dual-mode capability is increasingly important for the kind of layered agent architectures enterprises are building, where a planning agent hands off to specialist agents for execution.

The headline capability improvements in this version are in tool use and computer use, coding, and complex bug detection. Meta says the model can handle end-to-end agentic workflows — meaning it can receive a goal, break it into steps, use external tools to complete those steps, and return a finished result rather than just a response.

Pricing and Access

The Meta Model API launched in public preview for US developers on July 9. Pricing comes in at $1.25 per million input tokens and $4.25 per million output tokens. New accounts receive $20 in free credits to start with.

For context, that pricing is considerably cheaper than the GPT-5.6 Sol tier ($5 input / $30 output) that OpenAI launched the same day, though Sol sits at the frontier-reasoning end of the market. It also compares well against Anthropic’s Claude Opus 4.8 pricing for similar workloads.

EU availability is not yet confirmed. Meta has indicated European access is targeted for later this year, subject to regulatory review.

Why This Timing Matters

July 9 was a busy day for the AI model market. OpenAI launched GPT-5.6 with three tiers (Sol, Terra, Luna) alongside ChatGPT Work. Grok 4.5 from SpaceXAI became broadly available. Meta releasing Muse Spark 1.1 via API on the same day signals that this is no longer a two-horse race between OpenAI and Anthropic.

Businesses now have at least four credible frontier-tier model providers to evaluate: OpenAI, Anthropic, SpaceXAI, and Meta. That competition has direct implications for pricing and negotiating leverage that enterprises should pay attention to.

The fact that Meta is pricing its API aggressively — below GPT-5.6 Sol but with comparable multimodal and agentic capabilities — suggests Meta is prioritizing adoption over margin in the early months.

What This Means for Business

More options, lower prices. The rapid entry of Grok 4.5 and now Muse Spark 1.1 into the developer API market means the era of AI model price compression is accelerating. If you locked in long-term contracts with a single AI vendor in the last 12 months, now is a good time to revisit those terms.

Multimodal agents are table-stakes. Muse Spark 1.1’s ability to process images, video, and documents natively — not as an add-on — reflects where enterprise AI is heading. Use cases that previously required separate pipeline steps to extract information from documents or images can now be handled by a single model call.

The 1M-token context window changes what’s possible. A million tokens is enough context to load a company’s entire product documentation, a full year of customer support tickets, or a complete ERP data export. This isn’t theoretical — it’s the kind of scope that makes whole-business-context AI reasoning practical.

Multi-agent architecture support matters. The fact that Muse Spark 1.1 is explicitly designed to work inside multi-agent systems, both as an orchestrator and as a worker agent, signals where enterprise AI build-outs are headed. Point solutions are giving way to coordinated agent networks, and models that can participate on either end of that coordination will have an advantage.

For businesses evaluating AI infrastructure right now, the Meta Model API is worth a serious look — not to replace existing setups, but to benchmark against them and understand whether the price-performance case holds for specific workloads.

The AI model market just got more competitive. That is good news for any business building on it.