Meta opens Muse Spark 1.1 to developers
Meta released Muse Spark 1.1 on July 9, 2026, and opened a public preview of its new Meta Model API. The launch makes the model available to outside developers for the first time, while Muse Spark 1.1 is also becoming the engine behind Thinking mode in the Meta AI app and on meta.ai. The move is important because Meta is shifting from using its models mainly inside Facebook, Instagram, WhatsApp and its own assistant to competing directly for paid developer workloads.
Meta describes Muse Spark 1.1 as a multimodal reasoning model designed for agentic tasks. The company says it has improved tool use, computer use, coding and multimodal understanding. Its context window can hold up to one million tokens, and the model is designed to plan work, coordinate external tools and delegate parts of a task to subagents. For computer-based workflows, Meta says it can decide when a script is more efficient and when direct interaction with an interface makes more sense. It is also intended to debug large codebases, implement features and carry out code migrations.
The API is in public preview rather than general availability. Axios reports that Meta charges $1.25 per million input tokens and $4.25 per million output tokens. That is an aggressive entry point for a frontier-model provider and puts pressure on rivals to justify higher prices through quality, reliability or specialized features. Public API access does not make Muse Spark 1.1 an open-weight model; developers are buying hosted access from Meta.
For makers, the practical value is more choice for coding agents, research assistants and workflows that must keep track of long sessions. For businesses, the low published price may make pilots attractive, but cost per token is only one part of the decision. Teams still need to test accuracy, latency, tool reliability, data handling and how well the model recovers when an autonomous task goes wrong. A one-million-token window also does not guarantee that every detail will be used correctly.
Meta has published a separate evaluation report and says the model remains within its safety thresholds for chemical and biological risk, cybersecurity and loss of control. Those findings are self-reported and should be read alongside independent testing. The broader market signal is clearer: Meta is no longer relying only on distribution through billions of consumer accounts. It now wants developers and companies to choose its model as infrastructure.