Meta joins the AI price war! Launches its first paid large model with capabilities surpassing Google, priced at only a quarter of Anthropic's.
```
In the increasingly competitive AI tool market, Mark Zuckerberg hopes to win through pricing.
Meta Platforms has released its most advanced AI model, Muse Spark 1.1, and for the first time has launched a paid version for developers.
This is also the first time Meta has charged enterprises for model usage, opening up a new revenue source for the company. In an interview prior to the launch, Zuckerberg said the product will be one of the most competitively priced AI models on the market.
"Since this is not an open-source model, I think this is our first time seriously launching an API (application programming interface) service," Zuckerberg said:
"Our pricing will be very aggressive and very attractive."
Meta's stock opened lower and then rose on Thursday, up 0.19% at the time of writing.
Main feature is AI Agent capability, priced at 1/4 of competitors
Zuckerberg said, Muse Spark 1.1’s biggest improvement is in Agent (intelligent agent) capabilities.
Agents have become one of the hottest topics in the AI industry this year, usually referring to AI systems that can complete multi-step tasks on behalf of users. Zuckerberg said Muse Spark 1.1 has reached an “industry-leading or very close to leading” level in agent reasoning and tool invocation capabilities.
He also stated that the model has made significant advances in coding capabilities, and Meta employees have already used it to develop new products and features for various company apps.
At the same time, Meta will also launch a new Meta Model API platform to charge developers.
Zuckerberg revealed that the pricing for Meta API is about one quarter of the official price of top models from OpenAI and Anthropic. Developers can use the model for free up to a certain quota; fees are incurred only after exceeding the token allocation.
Zuckerberg said Meta's goal is to enable more people to use the company's AI technology.
"Some other AI labs' pricing is very high, with very high profit margins. We believe it is fully possible to provide cutting-edge, top-level intelligence at much lower costs."
Promises to invest hundreds of billions of dollars, Meta models surpass Google for the first time
42-year-old Zuckerberg is making major investments in AI to catch up with competitors like OpenAI and Alphabet, aiming for what he calls Superintelligence—AI that can outperform humans on various tasks.
Meta has pledged to invest hundreds of billions of dollars to build the infrastructure needed for superintelligence, including data centers and high-end AI chips. This week, the company also announced a $10 billion investment in data centers in Canada and unveiled a new image generation model.
Even so, Meta's previous generations of models have consistently lagged behind Anthropic, OpenAI, and Google in comprehensive tests.
However, Zuckerberg stated that Muse Spark 1.1's competitiveness has greatly improved, surpassing Google's Gemini model in agent capability, coding, and multimodal tests.
"This is a very important milestone. This is the first time Meta’s model has fully surpassed all Google models."
AI lab reorganization, strategic shift to closed-source paid models
Over the past year, Zuckerberg has made broad adjustments to Meta to advance its AI strategy.
After a disappointing model launch in spring 2025, Zuckerberg personally intervened to rebuild the AI team, hiring Scale AI founder Alexandr Wang to lead the new AI division, and conducting multiple rounds of layoffs and organizational restructuring.
Meanwhile, Meta's AI strategy has undergone a major shift.
Previously, Meta insisted on building free, open-source models; now, the company is prioritizing development of paid, closed-source models such as Muse Spark 1.1, a shift that almost rebuilds the entire tech stack.
Zuckerberg expressed satisfaction. "Overall, our progress is better than expected."
He admitted that Meta is still behind leading labs like Anthropic and OpenAI, but revealed that the company is developing a new model codenamed Watermelon, which may help Meta further push the boundaries of AI capabilities.
As for release timing, he said the main focus is still on improving model quality, so no specific schedule will be disclosed.
Zuckerberg: Mastering the base model is necessary to build the best AI assistant
Developing advanced AI models is extremely costly, but Zuckerberg sees this as aligning with Meta's long-term mission—to build a personal AI assistant that everyone can use.
He said that if Meta relies on other companies’ models, it can't guarantee those models are designed around the core scenario of a personal assistant.
"If you really want to create the best user experience, you must be able to shape the underlying technology." Zuckerberg said mastering the base model means Meta can "deliver what we think is the best product experience."
He also disagrees with the view that AI will ultimately be commoditized, with no difference between models.
Zuckerberg cited as an example Anthropic’s latest model Mythos, which is restricted due to US national security concerns, indicating that leading AI companies are already controlling the most advanced model capabilities rather than fully opening them up.
"Truly advanced capabilities have not spread widely," he said. "Anthropic kept its strongest model, only releasing a somewhat simplified version. Whatever their reason, at least it shows I don’t think the industry is bound to move towards all advanced AI capabilities being open to everyone."
Responding to commercialization doubts: Meta is building a complete AI profit model
Meta expects record capital expenditures in 2026, investing tens of billions of dollars to recruit AI talent and build the Meta Superintelligence Labs, and plans to continue investing hundreds of billions in AI infrastructure.
Such enormous investment has also raised external questions: how will Meta achieve profitability?
After announcing its earnings in April, despite beating expectations on revenue and profit, concerns over Meta’s lack of a clear AI business model led to a nearly 9% drop in its stock price.
In recent months, Meta has launched a series of commercialization measures, including:
Launching consumer AI chatbot subscription services;
Planning to sell AI Agents to enterprises;
Starting to charge for flagship AI model API access;
Planning cloud computing business, potentially selling computing power and AI infrastructure in the future.
Zuckerberg said Meta hopes that through more reasonable API pricing, more developers can access advanced AI technology. "Someone has to develop these models and ensure that the highest-quality intelligence benefits everyone," he said.
He emphasized that Muse Spark 1.1 will also power the free Meta AI chatbot, meaning ordinary users can still freely experience the latest model.
Risk Disclaimer & TermsThe market involves risk, investment needs caution. This article does not constitute personal investment advice, nor does it consider individual users' specific investment goals, financial conditions, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article suit their particular situation. Investment decisions based on this are at your own risk. ```