Meta spent years insisting the future of AI was open and free. Its whole strategy — the Llama line, the open weights, the “give it away and own the ecosystem” posture — was a deliberate contrast to OpenAI and Anthropic, which meter every token. On July 9, that posture cracked. Meta Superintelligence Labs shipped Muse Spark 1.1, a new agentic model, and wrapped it in something the company has never had before: a paid developer API.
The model is built for the job the whole industry is racing toward — autonomous, multi-step work rather than one-shot answers. Meta describes Muse Spark 1.1 as a multimodal reasoning model with a 1-million-token context window it actively manages, plus tool use, computer use across desktop, browser and mobile, and the ability to act as a lead agent that plans a project and hands execution to parallel subagents. Meta says it rivals GPT-5.5 and Anthropic’s Opus 4.8 on agentic benchmarks. Treat the benchmark claims with the usual caution — vendors grade their own homework — but the target is clear.
The real news is the storefront. The Meta Model API arrived alongside the model in public preview, priced at $1.25 per million input tokens and $4.25 per million output tokens, with $20 in free credits and US-only access to start. That is Meta’s first paid API, and it drops the company squarely into the same token-metering business it spent three years positioning against.
Our take: Forget the benchmarks — watch the business model. Open weights were Meta’s wedge: give the model away, own the developer ecosystem, deny rivals the mindshare that comes with it. A paid API is the opposite bet. It says Meta now believes there is real money in selling inference by the token, and that “free” wasn’t converting the agentic, enterprise workloads where the margin actually lives. Pricing it below the flagship US APIs is the tell — Meta isn’t leading on capability, it’s leading on cost, the same move Grok used to undercut Claude. The frontier is turning into a price war, and Meta just stepped behind a counter it swore it never would.
It also completes a picture we have been tracking. Meta is building its own inference silicon to cut what it pays to run models; a paid API is how it starts charging everyone else to run them. Own the chip, own the model, sell the tokens by the million — that is the playbook OpenAI wrote, and Meta is now running it.
What to watch
- Does the free credit convert? Twenty dollars gets developers in the door. The question is whether Muse Spark is sticky enough to keep them paying once the credits burn off — against incumbents with two years of tooling and habit on their side.
- US-only, for now. A preview fenced to the US is a soft launch. Watch how fast it opens up, and whether Europe trips over the same regulatory friction that has slowed other frontier rollouts.
- What happens to Llama. Meta hasn’t said open weights are dead. But every dollar of API revenue makes giving the next model away for free a harder internal argument.
- The price floor. With GPT-5.6 now public and a field full of cheap challengers, $1.25 / $4.25 may not stay the low bar for long.