Meta made a notable strategic move: Mark Zuckerberg announced Muse Spark 1.1, a 1-million-token-context agentic model — and, more significantly for developers, Meta's first-ever paid developer API, now in public preview.
What's in it
Muse Spark 1.1 is built for agentic work, not just chat. The headline capabilities:
- Computer use across desktop, browser, and mobile — the model can operate interfaces directly.
- Parallel subagent delegation — spinning up multiple sub-agents to work concurrently.
- 1M-token context — room for large tasks and long histories.
The public preview launched with $20 in free credits, US-only at first.
Why the paid API is the real news
Meta has historically been the champion of open models (the Llama line), released for anyone to run. A first-party paid API is a strategic shift — Meta competing directly with OpenAI, Google, and Anthropic on hosted, managed access, not just open weights. It signals that even the most open-leaning major lab sees value in offering a managed, monetized service alongside open releases.
Meta built its AI reputation on giving models away. A paid API says the strategy is now "both" — open weights and a hosted service.
The agentic angle
Muse Spark's focus on computer use and parallel subagents reflects where the whole field is pointed: from models that answer to agents that act. Computer use — operating apps by seeing the screen — remains hard and fragile (as we've covered), so the real test is reliability on actual tasks, not demo clips.
Confirmed vs. hype
Confirmed: the announcement, the paid-API preview, and the feature set as described by Meta. Independent evaluation of how well the computer-use and subagent features actually perform is the thing to watch — announcements describe intent; real-world reliability is earned. We'll follow up as developers put the preview through its paces.
