2026-07-16 11:03 UTCIn-site rewrite1 min readUpdated: 2026-07-16 13:49 UTC

The Sequence Opinion #896: Spark, Compute, and the Two Metas

Meta launched Muse Spark 1.1, the first Meta model with a price tag, marking a shift from open weights to a closed-source business model. As Meta builds a full vertical stack—from chips to cloud to apps—the question arises whether it can compete with frontier AI labs.

SourceTheSequenceAuthor: Jesus Rodriguez

Article intelligence

EngineersAdvanced

Key points

Meta released Muse Spark 1.1 with closed weights and paid API, priced at $1.25 per million input tokens and $4.25 per million output tokens, compatible with OpenAI endpoints.
Zuckerberg posted on X for the first time in three years to announce this strategic shift.
Meta also launched Muse Image and plans to build a cloud business called Meta Compute to sell surplus AI infrastructure.
Meta may have an edge at the app/agent layer, but faces structural challenges at the model-making layer.

Why it matters

This matters because meta released Muse Spark 1.1 with closed weights and paid API, priced at $1.25 per million input tokens and $4.25 per million output tokens, compatible with OpenAI endpoints.

Technical impact

May affect model selection, inference cost, product capability, and evaluation benchmarks.

This panel is AI-generated and reviewed for accuracy.

Last Thursday, Mark Zuckerberg posted on X for the first time in three years. That alone should tell you something. The occasion was the launch of Muse Spark 1.1, the second model out of Meta Superintelligence Labs and the first Meta model ever to ship with a price tag. It arrived with a public API, aggressive pricing at $1.25 per million input tokens and $4.25 per million output tokens, an OpenAI-compatible endpoint, and closed weights. Read that last part again. The company that spent three years evangelizing open weights as the moral and strategic high ground of AI just shipped a proprietary frontier model behind a paid API, and the CEO came back from a three-year social media exile to announce it.

Spark 1.1 did not arrive alone. Two days earlier Meta shipped Muse Image, its first image generation model from the new lab. A week before that, reports surfaced that Meta is building a cloud business, internally called Meta Compute, to sell surplus AI infrastructure to outside customers. Add the custom MTIA silicon ramping toward production and you get the picture: in roughly eighteen months, Meta has gone from an open-weights research shop with an ads business attached to a company assembling the entire vertical stack. Chips, datacenters, cloud, models, API, apps, devices. Only Google has ever held all of those cards at once.

So the question practically asks itself. Can Meta actually compete with the frontier labs? I think the honest answer is that this is two questions wearing one trench coat, and they have different answers. At the layer where models meet users, the app and agent layer, Meta might be the favorite. At the layer where models get made, the evidence is thin and the structural arguments cut against it. This essay argues both sides properly, because both sides deserve it.

The launch, read closely