AI News Feed
Market watch
LG UltraFine Evo 32U990A-S Hits Record-Low $1,099 on Amazon, TechRadar ReportsNewegg Labor Day Deal: 32GB Kingston Fury Beast DDR4-3200 Kit Nearly Half Price at $179.99 with CodeGalaxy Z Fold 8 inner screen corners reported with excessive giveHuawei unveils Mate XT 2 tri-fold with privacy-screen edition as Apple prepares folding iPhoneWall Street Analysts Recommend Nvidia, Uber and Marvell for Long-Term GrowthLenovo, AMD and Huawei Put Local Large-Model AI at Center of New Devices
LG UltraFine Evo 32U990A-S Hits Record-Low $1,099 on Amazon, TechRadar ReportsNewegg Labor Day Deal: 32GB Kingston Fury Beast DDR4-3200 Kit Nearly Half Price at $179.99 with CodeGalaxy Z Fold 8 inner screen corners reported with excessive giveHuawei unveils Mate XT 2 tri-fold with privacy-screen edition as Apple prepares folding iPhoneWall Street Analysts Recommend Nvidia, Uber and Marvell for Long-Term GrowthLenovo, AMD and Huawei Put Local Large-Model AI at Center of New Devices
Large Language Models

Meta releases Muse Spark 1.3, says it has caught up with Anthropic and OpenAI

Meta released Muse Spark 1.3 on Sept. 2, claiming it matches or beats Anthropic and OpenAI. Independent benchmarks put it behind only two Claude models.

AI News Feed · 2026-09-03 12:47

Meta said in a blog post that the large language model is available through its API to developers who pay for it, and will be rolled out to users of Facebook, Instagram and the Meta AI app in the coming days.

In an interview with Bloomberg, Meta Chief AI Officer Alexandr Wang said Muse Spark 1.3 represents the company's biggest performance jump so far and puts it at the same level as the latest models from OpenAI and Anthropic. Wang described the model as competitive with Anthropic's Claude Fable 5.1 and better than OpenAI's GPT-5.6 Sol, especially at generating code. He also claimed it outperforms "any of the current Chinese models out there."

Independent benchmarking is difficult to interpret, as models vary by task and vendors often optimize published tests. An evaluation by Artificial Analysis gave Muse Spark a score of 62 on its Intelligence Index, placing it behind only Claude Fable 5.1 and Claude Opus 5 and ahead of OpenAI's models. According to SiliconANGLE, Muse Spark 1.3 is the fourth version of the model family released in five months.

The release is part of Meta's effort to extract returns from multibillion-dollar AI investments. Last year, Chief Executive Mark Zuckerberg paid over $14 billion to buy a stake in ScaleAI, Alexandr Wang's former company, and hired him to lead Meta's Superintelligence Labs unit. Since then, Wang has worked to improve Meta's models and push out versions on an aggressive release cadence. But heavy spending has drawn investor scrutiny over when it will generate revenue, and Meta has shifted from the open-source approach of its Llama models to a proprietary strategy similar to Anthropic and OpenAI.

Meta started charging developers for access to Muse Spark 1.1 when it launched the model in July, with a price lower than rivals. Zuckerberg said the plan was to make the Muse Spark family one of the most affordable LLMs on the market. Wang said developers will pay no more for Muse Spark 1.3 than they did for Muse Spark 1.2, released in August. The model is available through Meta's Model API, along with tools for building AI applications.

Wang said the Muse Spark family is seeing rapid adoption, with some developers using "trillions of tokens per week." Muse Spark 1.3 uses roughly 25 percent fewer tokens for the same tasks than Muse Spark 1.2, supports multiple workflows in one session and handles longer, more complex instructions while retaining context. Wang added that it has greater awareness of its own limitations and is safer: when an action would be irreversible, it asks for confirmation first.

Meta has not yet decided whether to release the model's weights, even as Zuckerberg has authored a blog post stressing the need to make AI development more open and accessible. Open-weights models, popularized by Chinese AI makers, allow developers to run and modify models without paying, and Meta has previously promised to release model weights.