Meta officially enters AI model commercialization: Can low-price API strategy shake the dominance of OpenAI and Google?

On July 10, 2026, Mark Zuckerberg logged back into the X platform after three years of silence, posting a message strong enough to shake the entire AI industry: Meta officially launched the Muse Spark 1.1 multimodal reasoning model and, at the same time, opened a public preview of the Meta Model API. This tweet marks the official start of Meta’s transition from an “AI technology provider” to an “AI infrastructure services provider.”

This is not a routine product upgrade. Zuckerberg chose to announce the news on a competitor’s home turf—the X platform—rather than on Meta’s own social media ecosystem, and that in itself sent a clear signal. On the same day, the crypto market also rebounded: Bitcoin broke through $63,000, topping out at $63,925, with a 24-hour gain of 3.56%; Ethereum moved up in tandem to $1,772, up 2.66%. The total crypto market capitalization recovered to $2.19 trillion. On this day, the AI and crypto narrative lines intersected—one trying to open a path to commercialization through low-cost models, the other searching for direction as macro liquidity improved.

However, the capital markets’ response was rather restrained. As of Beijing time on July 10, Meta’s stock closed at $631.48, up 4.70% on the day. A 4.7% gain is not ordinary for a tech giant, but compared with the “breakout effect” expected from an AI launch, the market’s enthusiasm was clearly tempered. Investors no longer care about whether “Meta has AI,” but instead about whether “AI can make money.”

From Open Source to Paid: Why Meta Turned Around Now

To understand the deeper meaning of Meta’s strategic adjustment, it’s necessary to look back at how its AI path has evolved.

Over the past two years, Meta’s AI strategy has been centered on the core label of “open source.” From the ongoing open-sourcing of the Llama series models to building an AI research community, Meta tried to accumulate developer trust and industry influence through an open ecosystem. But this model has always faced a fundamental problem: open source cannot directly be converted into revenue.

After a model release in the spring of 2025 underperformed expectations, Zuckerberg personally stepped in to rebuild the AI team, hiring Scale AI founder Alexandr Wang to lead the newly established Meta Superintelligence Labs. The company’s strategy gradually shifted from “open source first” to developing “chargeable closed-source models.” Muse Spark 1.1 is the first concrete result of this strategic shift.

Meanwhile, Meta’s investment on the infrastructure side has reached an astonishing scale. In 2023, the company’s full-year capital expenditures were $28.1 billion; in 2024 they jumped to $39.2 billion; and in 2025 they reached $72.2 billion. In 2026, Meta sharply increased its full-year capital expenditures to $125 billion to $145 billion, focusing on AI compute clusters and large model R&D, with an investment scale about twice that of 2025. In just the first half of 2026 alone, Meta had already signed contracts for more than 5 GW of cloud computing and hosted data center compute resources.

With such massive infrastructure investment, there must be clear commercial outlets. The launch of Muse Spark 1.1 and the Meta Model API is, in essence, creating “a payback channel” for these hundreds of billions of dollars in capital expenditures.

The Differentiation of Muse Spark 1.1: Low Price Doesn’t Mean Low Specs

From a product perspective, Muse Spark 1.1 is not a rushed, on-the-spot piece of work. According to Meta’s official disclosure, it is a multimodal reasoning model built specifically for agent tasks, with clear enhancements across tool calling, computer operations, code generation, and multimodal understanding. The model supports a context window of 1 million Tokens, enabling it to serve as a lead agent for coordination tasks in multi-agent systems, or as a sub-agent to execute specialized work. As Zuckerberg revealed, Muse Spark 1.1 has already surpassed Google’s Gemini model in multiple test projects, including agent capabilities, programming, and multimodal aspects.

But what truly makes the industry sit up and take notice is Meta’s pricing strategy. The Meta Model API is priced at $1.25 per million input Tokens and $4.25 per million output Tokens. Zuckerberg said directly on X that this price is about one-quarter of the official prices for comparable top-tier models from OpenAI and Anthropic. Registered developers can also get a $20 free credit for trying it out.

It should be noted that this price is not “the absolute lowest.” It is higher than OpenAI’s entry-level GPT-5 mini and Anthropic’s low-cost Claude Haiku 4.5, but it is significantly lower than Anthropic’s high-end model Claude Sonnet 4.6. Meta’s pricing strategy targets the mid-to-high-end developer market—customers who need stronger model capabilities but are sensitive to the flagship product prices of OpenAI and Anthropic.

Four Titans, Four Paths

Comparing Meta with OpenAI, Anthropic, and Google on the same coordinate system makes it clear that there are four distinct commercialization logics.

OpenAI follows a “performance premium” route. Leveraging the technological lead of the GPT series, OpenAI charges high API fees to enterprise customers while distributing model capabilities through Microsoft’s cloud channels. Its core assumption is that as long as the model is strong enough, enterprises are willing to pay a premium for performance.

Anthropic bets on a “safety premium.” With “Constitutional AI” and safety as its differentiating labels, Anthropic has attracted a large number of enterprise clients with high requirements for compliance and risk control. Its valuation has soared to $1.2 trillion in the secondary market, reflecting the capital market’s recognition of the commercial value of “safe AI.”

Google adopts a “full ecosystem integration” strategy. Gemini models are embedded across Google’s entire product suite—search, advertising, cloud, Workspace, and more—so AI capability becomes a tool to enhance the ARPU value of existing businesses rather than a standalone source of revenue.

Meta chooses the fourth path: open ecosystem + cost advantage. By attracting developers to join at scale with API prices far below those of competitors, Meta uses ecosystem scale to counter OpenAI’s technical moat and Google’s ecosystem moat. The logic chain is: lower price → more developers use it → larger ecosystem scale → data flywheel and network effects → formation of long-term competitive advantages.

None of these four paths is absolutely better or worse, but Meta’s strategy has one standout characteristic: it does not rely on a technological gap to win; instead, it tries to rebuild the foundation of competition with an economic model. If the capability gap among AI models continues to narrow over the next 12–24 months, price will become a higher-weight variable in enterprise decision-making—this is exactly the core assumption Meta is betting on.

Why the Market Didn’t “All In”

After the news was announced, Meta’s stock rose 4.7%, closing at $631.48. While this gain would be impressive in any ordinary product launch, given that Muse Spark 1.1 is Meta’s first strategic product to charge enterprises for model access and open up a brand-new revenue stream, the market’s reaction can only be described as “cautiously optimistic.”

Investors are not discounting Meta’s AI capabilities; rather, they are focusing on three deeper issues.

First, the certainty of revenue contribution. Since its API pricing is only one-quarter of competitors, Meta would need to achieve several times the rivals’ call volumes at scale to generate revenue at the same level. Muse Spark 1.1 is currently only open for public preview to U.S. developers. From preview to large-scale commercial use, and then to meaningful revenue contribution, there is still a long road ahead.

Second, the sustainability of capital expenditures. With annual capital expenditures of $125 billion to $145 billion, Meta is burning more than $340 million per day on AI infrastructure. Even if Meta’s advertising business continues to grow—WARC Media predicts its 2026 ad revenue will reach $240 billion—such massive spending still puts continuous pressure on the income statement.

Third, the length of the profitability cycle. AI infrastructure investment takes time to translate into profits. Goldman Sachs predicts that the combined 2026 capital expenditures of Alphabet, Amazon, Microsoft, and Meta will reach $725 billion. Such enormous industry-wide investment means the AI commercialization story is not something that can be realized in just one or two quarters.

The market has moved from the “AI story” phase into the “AI realization” phase. Investors are no longer paying for the model “release” itself; they need to see how the model turns into cash flow.

Conclusion

On the day Zuckerberg returned to X, Meta used Muse Spark 1.1 and the Model API to send a clear message to the industry: the AI race is shifting from “who has the better model” to “who can make the model available to more people at lower cost.”

OpenAI has technological moats, Google has ecosystem moats, and Anthropic has safety moats—Meta is choosing to pry open the market using a price moat. Whether this path can work depends on two premises: first, whether the gap in model capabilities is truly narrowing; second, whether developers will actually migrate because of price.

For the crypto industry, regardless of the outcome of this competition, lower-cost AI infrastructure means more possibilities. When model calls are no longer a cost bottleneck, the imagination space for on-chain intelligent applications will be redefined.

The commercial story of AI has only just turned to its second chapter. The first chapter was “who built the model”; the second chapter is “who can make the model affordable for others.” Meta is throwing its full effort behind writing the second chapter.

FAQ

Q1: What is the exact pricing of the Meta Model API? What advantages does it have compared with competitors?

The Meta Model API is priced at $1.25 per million input Tokens and $4.25 per million output Tokens. Zuckerberg said this price is about one-quarter of the official prices for top-tier models from OpenAI and Anthropic. Registered developers can also receive a $20 free trial credit.

Q2: What are the core capabilities of Muse Spark 1.1?

Muse Spark 1.1 is a multimodal reasoning model built specifically for agent tasks, with clear enhancements in tool calling, computer operations, code generation, and multimodal understanding. The model supports a context window of 1 million Tokens and can serve as a lead agent coordinating tasks in multi-agent systems or as a sub-agent to execute specialized work.

Q3: Why did Meta shift from open-source Llama to a paid API model?

Meta’s investment in AI infrastructure has reached $125 billion to $145 billion per year, and the open-source model cannot provide commercial returns for such huge spending. Switching to a paid API is to look for a sustainable payback channel for hundreds of billions of dollars in AI capital expenditures, while also using a low-price strategy to attract developers and build ecosystem scale.

Q4: Why did Meta’s stock rise by only 4.7% after the AI release?

Investors’ focus has shifted from “releasing an AI model” to “whether AI commercialization capabilities can translate into real revenue.” The market’s doubts about Meta center on three areas: the certainty of API revenue contribution, the sustainability of capital expenditures at the $125 billion scale, and the time cycle for AI investment to turn into profits.