Meta Releases Muse Spark Open, and Challenges China

Meta has released Muse Spark Open, an open-weight AI model that could challenge frontrunners and boost Meta’s AI position.

China currently dominates the open-weight market, with models such as Kimi K3 and GLM 5.5 among the most popular. However, Meta’s new model could give the West a significant leg-up.

Open-weight models are cheaper and can be run locally, which brings serious security benefits. As AI becomes more of a security concern, the models winning in the open-weight category could prove deciders in the wider AI race.

Meta has turned its attention to open-weight AI models with the release of Muse Spark Open, in a collaboration with Apple that sees the model able to run locally on Mac Studios.

Alexandr Wang, Meta’s chief of AI, said in a post on X that the new model is “the most capable US open-weight model to date.” Evaluation results released by the company on Thursday do suggest the model offers superior performance.

This just in! View the top business tech deals for 2026 👨‍💻

According to Artificial Analysis’s Intelligence Index, Muse Spark Open sits ahead of several Western open-weight models, achieving a score of 39. Similarly, Spark Open either matches or tops rivals on multimodal and knowledge tasks.

Artificial Analysis’s Intelligence Index might have placed Muse Spark Open above some Western models, but it remains behind Z.ai’s GLM 5.5 and Moonshot AI’s Kimi K3, both of which are Chinese-owned.

Z.ai and Moonshot AI, as well as DeepSeek, have been leading the charge when it comes to open-weight models. Mostly, they’ve been able to stay competitive with frontier labs by offering lower prices and maintaining solid performance metrics.

While Western models from OpenAI and Anthropic generally perform better than Chinese models, it’s widely believed the gap is closing. The concern is so great for the US government that it has discouraged US businesses from using Chinese AI in the past.

With the arrival of Muse Spark Open, however, the US might have a way to challenge Chinese open-weight authority. The new model’s release might also re-establish Meta in the US AI race, which has been dominated by the likes of Anthropic and OpenAI.

Days before Spark Open was announced, Meta introduced Muse Wonder, which it said rivals the most advanced version of ChatGPT and Claude. Indeed, Artificial Analysis’s Intelligence Index placed Muse Wonder ahead of OpenAI’s GPT-Astra, Claude’s Fable 5, and Google’s Gemini 4.

There are three types of AI models: closed, open-weight, and open-source. Of the three, open-source provides the most transparency, by allowing comprehensive access to their components and even allowing users to retrain models.

Open-weight models let developers download their weights, or their parameters, and modify them. They don’t allow for the same accessibility as open-source models, but they offer more than closed models, which aren’t downloadable and keep much of their architecture hidden.

Closed models make up today’s highest performing AI models. Think Claude Opus 5.5 or GPT-6. However, open-weight models are becoming increasingly popular for today’s businesses, due to clear advantages.

The first advantage is price. Open-weight models are generally cheaper, with lower token costs, because they’re run by multiple hosts rather than by a single provider.

Furthermore, open-weight models are stored locally on proprietary infrastructure, so data never leaves a company’s servers. This provides a security and privacy advantage, particularly for those working in industries that require confidential or sensitive customer data. Given the current climate of AI data breaches and cyber attacks, this is a big plus.

As the security concerns around AI become more significant, businesses may increasingly turn to open-weight models to better protect themselves and their data. This could lead to open-weight models becoming the preference over closed models.

It’s important to note that closed models do currently hold the edge when it comes to performance, especially around complex, multi-step reasoning. But, this gap could certainly change, and the US staking a claim in this currently Chinese-dominated market isn’t a bad idea.

Nicole is Tech.co’s News Editor, reporting on the latest technology news and curating The AI Strat newsletter. After studying English Literature and Creative Writing, they worked on local newspapers and online publications, including Outlander Magazine. Previously, they covered tech products and news at Expert Reviews. Outside of Tech.co, they enjoy sports and video games.