Meta releases a more powerful AI model, claiming it has further narrowed the gap with its competitors.

Meta released its most powerful AI model to date, Muse Spark 1.3, on Wednesday. Chief AI Officer Alexandr Wang said its performance has caught up with or even surpassed that of major competitors, marking a significant step forward for the social media giant in the AI arms race.

In an interview, Wang stated that Muse Spark 1.3 is "superior" to OpenAI's GPT-5.6 Sol in terms of programming capabilities, and "on par" with Anthropic's Claude Fable 5.1.

He characterized this update as Meta's "biggest leap in model performance to date."

The model will be made available to developers for a fee and will be gradually rolled out to users of Meta AI's social media platforms such as Instagram and Facebook. Meta also revealed that development of its more anticipated large-scale model, "Watermelon," is progressing as planned.

Performance leap, facing top-tier competitors

Muse Spark 1.3 has been upgraded in several key metrics:

Improved operational efficiency reduces the number of tokens required by 25%;Supports advancing multiple workflows simultaneously without requiring separate sessions;The ability to handle long and complex instructions and the ability to retain details across tasks have both been improved;At the same time, it has a stronger sense of self-limitation, and its security is also enhanced.

Wang specifically pointed out that the model's proxy capabilities (i.e., its ability to complete tasks autonomously on behalf of the user) have been significantly improved. However, he also acknowledged that OpenAI is about to launch a more advanced new model, "Astra," and the competitive landscape will continue to evolve.

It is worth noting that horizontal comparisons between AI models have inherent limitations—different models perform differently on different tasks, and evaluation benchmarks can be optimized and adjusted, which may not accurately reflect the actual user experience.

Commercialization is accelerating, and the developer ecosystem is expanding rapidly.

Muse Spark 1.3 maintains the same pricing structure for developers as its predecessor, Muse Spark 1.2, and access is provided through the Meta Model API platform.

Wang revealed that the API platform has achieved strong market adoption, with some developers using "trillions of tokens per week".

This is Meta's latest move to commercialize AI. In July of this year, Meta began charging developers for the first time for using Muse Spark 1.1, with CEO Mark Zuckerberg stating at the time that the pricing goal was to be one of the most competitive options on the market.

At the same time, Meta is also working on building a cloud infrastructure business, selling AI computing power and model access permissions to external parties.

The dilemma of balancing openness and profitability

Regarding its open-source strategy, Meta has not yet decided whether to release the weight file for Muse Spark 1.3—the internal blueprint formed during model training that determines how the model responds.

Wang stated that the company still plans to release weights for the previous version, Muse Spark 1.2.

Zuckerberg recently wrote an article elaborating on the importance of open AI development, but how to strike a balance between the open philosophy and the expected return on huge AI investments remains a core issue facing Meta.

Meta is betting hundreds of billions of dollars on the AI sector, and the massive spending has raised questions among investors about the return on investment.

Security risks: Early models had infiltrated external systems.

This release is also accompanied by background pressures related to security.

Wang revealed that an early Meta model accessed the internet and hacked into external service systems during cybersecurity testing, echoing similar incidents that have recently occurred at other model companies, exacerbating concerns about the controllability of AI technology.

Wang stated that this incident provides an important reference for Meta to strengthen the security of Muse Spark 1.3. He said:

Before deciding to release it, we conducted comprehensive and in-depth security testing and training on this model.

The new model will proactively seek confirmation before performing irreversible operations and request clarification from users when necessary to reduce potential risks.

Looking ahead to the longer term, Wang said that the development of Meta's highly anticipated large-scale model "Watermelon" is progressing as expected, and that "Watermelon will be extremely competitive," but did not disclose a specific release timeline.

To accelerate the narrowing of the technology gap with competitors, Zuckerberg overhauled the company’s AI strategy last year, poaching Wang from his company and appointing him to lead the newly formed Meta Superintelligence Labs (MSL).

Since taking office, Wang has consistently launched new models and products at a steady pace.

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.