Meta turns Muse Spark into its paid AI bet

234

TL;DR:

Mark Zuckerberg returned to X after more than three years to announce Muse Spark 1.1, Meta’s new AI model for coding, agentic tasks and multimodal reasoning. The bigger move is not the post. It is Meta opening a paid Model API, giving developers access to Muse Spark and pushing the company deeper into the enterprise AI market.

Zuckerberg described Muse Spark 1.1 as “a strong agentic and coding model at a very low price”. Meta says the model can manage a 1 million-token context window, work across tools and computer interfaces, and handle coding, images, video and long workflows.

The pricing makes the launch sharper. Reuters reported that US developers can test the public preview with $20 in credits before usage-based pricing of $1.25 per million input tokens and $4.25 per million output tokens. That positions Meta directly against OpenAI and Anthropic on cost, not just capability.

Meta is also trying to shift perception. Its Llama strategy made it influential in open models, but Muse Spark 1.1 is a hosted, metered product. That matters because AI costs are rising, and investors want proof that Meta can turn infrastructure spending into revenue beyond advertising.

Early partners are already framing the model as useful for real developer work. Replit CEO Amjad Masad called Muse Spark “a complete agentic foundation”, while Meta says the model improves on coding, tool use and computer-use workflows.

The next test is adoption. Cheap pricing may win trials. Sustained developer trust will depend on reliability, transparency and whether Muse Spark performs outside Meta’s own benchmarks.