Meta released Muse Spark 1.1 on Thursday, its latest artificial intelligence model designed for coding and agentic work, according to the company's chief AI officer Alexandr Wang. The update comes three months after Meta unveiled its first AI model under Wang's leadership and represents the company's effort to compete with OpenAI and Anthropic in the AI coding market. Wang told CNBC the model is Meta's "strongest model for agentic and coding work yet." The release reflects growing Wall Street pressure on Meta CEO Mark Zuckerberg to demonstrate returns on the company's massive AI infrastructure investments, as Meta has failed to keep pace with competitors in developing popular AI models and applications despite spending at rates comparable to its hyperscaler peers.
Meta Sets Aggressive Pricing for Muse Spark 1.1 API Access
Meta is making the new model's API available through a developer portal as part of a public preview. A Meta spokesperson said some early partners can already access the API, while new users "will be able to add themselves to a waitlist and be added from there over time." The company is limiting API access to its own properties rather than making it available on third-party platforms like the OpenRouter marketplace.
Wang characterized pricing as "very aggressive and attractive" compared with similar offerings from Anthropic and OpenAI. Every new API account will start with $20 in free credits. Meta will charge $1.25 per million tokens in input and $4.25 per million tokens of output. "The goal is to really have attractive pricing that scales with immense consumption usage," Wang said.
The initial Muse Spark model released in April was only available to "select partners" who could access the technology via a "private API preview."
Muse Spark 1.1 Targets Coding and Agent Capabilities
Wang said Muse Spark 1.1 outperformed rival models in certain tasks involving the ability to interact with various third-party coding products and services. Meta's Superintelligence Labs trained Muse Spark 1.1 to excel in coding-related tasks because that ultimately improves the capabilities of AI agents that can autonomously perform multiple tasks, Wang said.
"You kind of have to build coding capabilities as part of that in service of overall agentic capabilities," Wang said. He added that Meta trained Muse Spark 1.1 "to be able to work well with all of the most popular harnesses that developers use today, and we felt that was the best approach for this model given our goal to maximize adoption."
The release is Meta's second notable rollout for the Muse family this week. On Tuesday, Meta released Muse Image, originally code-named Mango, a model for creating images.
Meta Continues Proprietary AI Development Alongside Open Source Commitment
Although Meta's previous AI strategy emphasized releasing its earlier Llama family of models to the open-source community, the company is now focusing on selling access to proprietary AI models. Wang said Meta is still "committed to open source" and that his unit has a "variant of Muse Spark that is in development that we do intend to open source." He declined to say when the company would release it.
Wang said Meta is currently training a more powerful AI model, code-named Watermelon, but didn't say when it would be released. Muse Spark's code name was Avocado.
FAQ
What did Meta release on Thursday?
Meta released Muse Spark 1.1 on Thursday, an AI model designed for coding and agentic work. According to chief AI officer Alexandr Wang, it represents Meta's "strongest model for agentic and coding work yet."
How much does Meta charge for Muse Spark 1.1 API access?
Meta provides $20 in free credits for every new API account. After that, the company charges $1.25 per million tokens in input and $4.25 per million tokens of output. Wang described the pricing as "very aggressive and attractive" compared with similar offerings from competitors.
Why did Meta develop Muse Spark 1.1 for coding tasks?
Meta trained Muse Spark 1.1 to excel in coding-related tasks because coding capabilities improve the performance of AI agents that can autonomously perform multiple tasks. Wang said "you kind of have to build coding capabilities as part of that in service of overall agentic capabilities."
