Meta just "pricemogged" OpenAI and Anthropic with 75% cheaper AI

Signals Inbox·July 13, 2026·Frontier AI

Meta has launched its first paid model API with a blunt pitch: frontier-level coding and agent work should not cost this much. Muse Spark 1.1 starts at $1.25 per million input tokens and $4.25 per million output tokens, roughly one-quarter of what Meta says comparable OpenAI and Anthropic models cost. The real signal is not another model launch. It is Meta using its ad-funded compute empire to attack the margins of AI labs that need model revenue to survive.

Q1What did Meta actually launch?

Meta launched paid developer access to Muse Spark 1.1 through its new Model API. According to Meta AI chief Alexandr Wang, the model costs $1.25 per million input tokens and $4.25 per million output tokens. It is Meta’s strongest model yet for coding and agentic work, meaning tasks where AI writes software, uses tools, and completes several steps on its own.

Q2Where does the 75% cheaper claim come from?

Meta says its rates are around one-quarter of what OpenAI and Anthropic charge for comparable high-end models. That is another way of saying roughly 75% cheaper. The comparison is not perfect because every model has different speed, accuracy, and token use. But the list price is aggressive enough to force developers to ask why they are paying several times more elsewhere.

Q3Why can Meta afford to price this low?

Because Meta does not need the API to fund the entire company. OpenAI and Anthropic depend heavily on selling models and subscriptions. Meta generated its core business from advertising long before this API existed. It can use that cash, plus the huge computing network already built for Facebook, Instagram, WhatsApp, and Meta AI, to accept lower model margins while it pulls developers into its ecosystem.

Q4Is Muse Spark really as good as the expensive models?

Not across everything. Meta’s evaluations show Muse Spark competing well on coding, tool use, reasoning, and healthcare tasks, but it does not clearly beat every frontier rival on raw intelligence. That makes the value argument more important than the benchmark crown. A model can lose a few tests and still win customers if it completes everyday coding work reliably at one-quarter of the price.

Q5Does 75% cheaper mean every AI bill falls 75%?

No. Models do not always use the same number of tokens to solve a task. A cheaper model can think longer, retry more often, or produce weaker answers that need extra checking. One recent study found that lower list prices sometimes produced higher final costs because reasoning-token use varied so much. Developers will need to compare the cost of a finished task, not just the price printed beside one million tokens.

Q6So what changes now?

OpenAI and Anthropic now have a credible price problem. Meta is entering their market after spending years giving models away, and it is starting with rates designed to make premium margins look excessive. If Muse Spark proves reliable in real coding agents, competitors may have to cut prices, offer bigger volume discounts, or explain why their models are worth four times more. The AI race is moving from who has the smartest demo to who can deliver useful work at the lowest cost.