Meta Platforms Launches Muse Spark 1.3 with Enhanced AI Performance

On September 2, 2026, Meta Platforms unveiled Muse Spark 1.3, marking the third major update to its AI model family in just five months. This new version introduces a 20% reduction in tool calls and a 25% decrease in token consumption compared to its predecessor, Muse Spark 1.2, enhancing efficiency for developers.

The core upgrades of Muse Spark 1.3 focus on agentic workflows and coding performance. It has been trained on a wider range of long-horizon coding tasks, allowing it to perform better with fewer external tool calls, which can introduce latency and errors. The reduction in token usage is particularly significant, as tokens are the billing unit for most AI APIs, translating directly to cost savings for users.

Muse Spark 1.3 is now available through Muse Code and the Meta Model API, with plans for a “max reasoning” mode pending further safety testing. The rapid development of the Muse Spark family, which began in April 2026, reflects Meta’s strategic pivot from its open-source Llama models to a more proprietary and monetized approach.

Alexandr Wang, Meta’s chief AI officer, noted that Muse Spark's capabilities are increasingly competitive with leading models from OpenAI and Anthropic. This shift signifies Meta’s ambition to create a “personal superintelligence” experience, where AI not only answers questions but also collaborates on complex tasks.

As Meta continues to invest heavily in AI infrastructure, with projected spending between $130 billion and $145 billion in 2026, the company aims to solidify its position in the competitive AI landscape.