Meta Launches First Paid AI Model Muse Spark 1.1, Entering the Coding Market
N.R. Finch
Meta on July 9 began charging for AI model API access for the first time, releasing the coding-focused Muse Spark 1.1 at $4.25 per million output tokens — a concrete shift from open-source to closed-source commercialization, aimed squarely at the paid API market led by OpenAI and Anthropic.
Is this pricing cheap or expensive?
Input costs $1.25 per million tokens; output costs $4.25 per million tokens — above OpenAI's entry-level GPT-5 mini and Anthropic's budget Claude Haiku 4.5, but below the premium Claude Sonnet 4.6.
Zuckerberg said the pricing is roughly 25% of other top models' list prices, calling competitors' margins "extremely high." This means → Meta is buying market share with low prices, not chasing per-unit profit.
New users get a $20 free credit for testing, then pay per use. Access is US-only for now; developers apply through Meta's own portal and join a waitlist.
Why lead with coding, not chat or images?
Muse Spark 1.1 was trained by Meta Superintelligence Labs (MSL — Meta's dedicated frontier-AI research unit) and positioned as Meta's "strongest model for agentic and coding tasks."
In plain terms = coding ability is the foundation for building agents — AI systems that call tools and complete multi-step tasks on their own. Master code first, then the AI can actually do work.
The model has been specifically optimized for the mainstream tool frameworks developers use most. This reflects Meta prioritizing "will developers actually adopt it" over "highest benchmark score."
Where does Meta actually rank against Google, OpenAI?
Zuckerberg said Muse Spark 1.1 outperforms Google's Gemini across agentic reasoning, tool use, and multimodal benchmarks, adding that "this may be the first time a Meta model has comprehensively surpassed all of Google's models."
He also acknowledged Meta still trails Anthropic and OpenAI. This means → Meta's competitive position is "caught up with Google, not yet caught up with the leaders" — a second-tier player pushing into the first tier.
Put simply = this launch is Meta proving to the market "we can build a paid-grade product," not declaring victory.
Why is Meta's AI strategy pivoting now?
Meta shipped two major AI products this week: on Tuesday, the image-generation model Muse Image (internal codename Mango) for creators and advertisers; on Wednesday, the coding model Muse Spark 1.1 for developers.
Muse Spark 1.1 is expected to gradually replace the Llama open-source models currently running on WhatsApp, Instagram, Facebook, and Meta's smart glasses. This means → Meta is not abandoning open source — it is layering a closed-source paid product line on top of it.
Meta is also training a more powerful model codenamed "Watermelon," though no timeline has been disclosed. Zuckerberg stressed that the current focus is quality, not speed.
What does this mean for the market?
Meta charging for model API access for the first time puts it directly into the paid market dominated by OpenAI and Anthropic — this is a strategic validation point, not just a product launch.
In plain terms = if the low-price strategy attracts enough developers, Meta proves that "an open-source giant can also make money from closed-source." If it doesn't, cheapness alone isn't enough — the model itself has to be strong enough.
This signals the AI industry is shifting from "who has the strongest model" to "who can actually sell a model" — commercialization capability is starting to matter as much as technical capability.
Content is for reference only, not financial advice.