Mark Zuckerberg spent the day before the launch telling anyone who would listen that Meta’s new model would be the cheapest serious option on the market. On July 10, the company made the claim official: Meta released Muse Spark 1.1, its most advanced AI model, and for the first time began charging developers for the privilege of using it.
The move opens a new revenue line for a company that has given away its AI models since the Llama releases. Spark 1.1 arrives with a paid tier for developers and a new billing system, the Meta Model API, built to charge for usage. The headline number is the price: Zuckerberg said API pricing will run at roughly 25 percent of what OpenAI and Anthropic charge for their top-tier models.
Developers can still use Meta’s models free of charge, but only up to a token threshold. Beyond that ceiling, usage bills begin. The structure is a deliberate bridge between Meta’s open-source history and its new commercial ambitions, and Zuckerberg framed it as a market intervention. “Some other AI labs price very high and run very high margins,” he said in a pre-launch interview. “Our strategy is to get as many people using our AI as possible. We think we can deliver frontier-level, or at least high-level, AI at a much lower cost.”
The model itself extends Meta’s reach beyond code. Spark 1.1 can generate software, but it can also simulate a user’s actions, clicking through applications, filling out forms, navigating interfaces, which pushes AI agent capabilities into settings that have nothing to do with programming. Computerworld, reporting on the release amid a corporate cost review of AI spending, noted that Spark 1.1 is priced well below GitHub Copilot and Cursor, the two dominant tools in the AI coding market.
Meta said its training costs for the model ran about one-fifth of what competitors spent on comparable systems. The claim, which Meta did not break down in detail, points to the company’s broader argument: that its computing infrastructure, built at enormous scale for social media, gives it a structural cost advantage in training and serving models.
The pricing decision lands at a delicate moment for the enterprise AI market. Corporate buyers have begun scrutinizing AI budgets after two years of rapid deployment, and several large companies have told vendors they will consolidate spending on a smaller number of tools. Into that environment comes Meta, offering a frontier-adjacent model at a quarter of the going rate. “Price is now the battlefield,” one analyst who covers developer tools said. “The incumbents are going to have to answer.”
GitHub and OpenAI have responded before, offering cheaper tiers and usage-based plans as competition intensified. The difference this time is the scale of the discount and the identity of the discounting company: Meta is not a niche challenger but one of the largest computing companies in the world, with the data centers and distribution to serve enterprise customers at volume.
The pricing also positions Meta against the open-weight movement. Google has been distributing capable small models free of charge, and the open-source community has grown increasingly adept at fine-tuning existing weights into useful products. Meta’s answer is to straddle both worlds: the free tier keeps its models inside the open-weights conversation, while the paid API captures the developers who need hosted scale, reliability and support. The strategy, analysts said, is to let the community market the model while the API harvests the revenue.
The free tier does double duty. It lets Meta keep its developer ecosystem open and its models in the hands of the widest possible audience, the strategy that made Llama the most downloaded open model family. At the same time, the token threshold creates a conversion funnel: small experiments stay free, and real workloads begin to pay. Analysts said the threshold is where Meta’s revenue story will be decided, because usage-based billing converts at whatever rate the market’s existing habits allow.
Zuckerberg’s framing of high-margin competitors was pointed. OpenAI and Anthropic have built their businesses on premium pricing, arguing that frontier capability justifies the cost. Meta’s counter-argument, that capable AI can be delivered cheaply at scale, is a direct attack on that premise, and it puts pressure on the entire pricing structure of the industry’s top-tier model market.
For investors, the launch is Meta’s clearest signal yet that it intends to monetize its AI work directly rather than only through improved ads. The company’s ad business remains the engine, but a paid model API with a 25 percent price advantage is a test of whether Meta can build a second business on top of it. The first quarter of billing data will be the real answer, and developers who have been waiting for a cheaper frontier option will be watching the token threshold, and their invoices, closely.
