Meta’s Superintelligence Labs shipped Muse Spark 1.1 on July 9. It’s Meta’s move into the AI coding and agentic arena, aimed squarely at Anthropic and OpenAI. And for the first time ever, Meta is charging money for a model.
What it actually is
A multimodal reasoning model built for agent work. It plans multi-step tasks, calls tools, drives a computer, writes and debugs code across big messy codebases, and reads images, video, audio and PDFs to feed all of that. As the lead agent it makes a plan and farms work out to parallel subagents; as a subagent it stays in its lane and escalates when stuck. Meta’s own benchmarks tell the honest story: it tops tool-use and orchestration boards, but trails Opus 4.8 and GPT-5.5 on pure coding.
The API is the real news
The Meta Model API is in public preview for U.S. developers, and it speaks both the OpenAI SDK and Anthropic’s Messages format — so swapping it into an existing agent stack is close to a config change. New accounts get $20 in credits, then pay-as-you-go: $1.25 per million input tokens, $4.25 per million output. It’s already live in the Meta AI app’s Thinking mode.
You Might Also Like
- Cursor Composer 2 Takes on Anthropic and Openai With a 0 50 m Token Coding Model and the Benchmarks Back it up
- Fara 7b Microsofts Tiny Model That can Actually use Your Computer
- Openai Oauth Turns Your Chatgpt Subscription Into a Free Openai api but Should you use it
- Anthropic Gives Claude Computer use on mac and you can Control it From Your Phone
- Meta Muse Spark 14b and 9 Months Later Alexandr Wang Delivers but its Closed Source
Top AI Product 
Leave a comment