Meta Superintelligence Labs (MSL) officially released Muse Spark 1.1 on July 9; Meta AI head Alexandr Wang told CNBC that Muse Spark 1.1 is currently Meta's "most powerful agent task and coding work model."
Muse Spark 1.1 Core Specifications: Around 1 Million Token Context
According to Meta's official statement, the main specifications and features of Muse Spark 1.1 are as follows:
Context Window: 1 million tokens, capable of easily retrieving historical records, actively remembering, and compressing key steps.
Multi-Agent Collaboration: Can act as a "main agent" overseeing the overall situation (gathering context, planning, delegating sub-agents) or focus on a single responsibility as a "sub-agent" (intelligent judgment on when to escalate and report).
Computer Use: Executes dynamic workflows across multiple applications, navigating unfamiliar UIs with minimal human intervention; example: when organizing a dinner and encountering sudden changes in reservations, it can proactively detect and update decisions in real-time.
Advanced Coding: Diagnosing complex bugs, implementing enterprise-level new features, executing large-scale code migrations; widely used by Meta engineers and researchers.
Multimodal Vision: Converts visual input into code, generates highly detailed image and video descriptions; Facebook Marketplace example: automatically extracting product photos and listing items using only a mobile video.
Pricing Strategy: Input $1.25, Output $4.25 per Million Tokens
According to Alexandr Wang on CNBC, the API pricing for Muse Spark 1.1 is $1.25 per million tokens for input and $4.25 per million tokens for output; each new API account receives a free credit of $20.
Wang states that the pricing is "very aggressive and attractive" compared to similar products from labs like Anthropic and OpenAI, aiming to maximize adoption to support large-scale usage growth.
Currently, the API is only available in a public preview on Meta's developer portal and has not been opened to third-party platforms like OpenRouter; new users can join a waitlist to gradually gain access.
Meta Shifts from Open Source to Proprietary Commercialization, Watermelon Still in Training
Reports indicate that Meta previously focused its AI strategy on releasing Llama series models to the open-source community, but is now more focused on selling access to proprietary AI models; Wang said Meta remains "committed to open source," and MSL is developing an open-source variant of Muse Spark, though no release date has been announced.
Wang also revealed that Meta is training a more powerful model codenamed "Watermelon" (Muse Spark's internal codename is Avocado), but no release timeline was provided.
This week, Meta also released Muse Image (an image generation model, previously codenamed Mango), targeting creators and advertisers.
Frequently Asked Questions
What is the API pricing for Muse Spark 1.1, and how can I access it?
According to Wang, the pricing is $1.25 per million tokens for input and $4.25 per million tokens for output, with a $20 free credit for new API accounts. Developers can access it via the public preview on Meta's developer portal, and new users can join a waitlist. General users can use it in the "Thinking" mode of Meta AI App and on meta.ai website; it is not yet available on third-party platforms like OpenRouter.
How does Muse Spark 1.1 differ from the original Muse Spark?
The original Muse Spark launched in April 2026, offering private API previews only to select partners. Muse Spark 1.1 has been further opened as a public preview in the developer portal, with enhancements in agent collaboration, computer control, and coding capabilities, supporting a context window of around 1 million tokens.
What is Meta's next more powerful AI model?
Wang disclosed that Meta is training a more powerful model codenamed "Watermelon," but no release date has been given. An open-source variant of Muse Spark is also planned, but no specific timeline has been announced.
