Meta's Muse Spark 1.1: Outperforms Gemini in Coding & Reasoning

Meta has released a new version of its AI model, Muse Spark, which it claims surpasses previous products from OpenAI, Anthropic, and Google. The new model, called Muse Spark 1.1, can handle tasks like coding, video captioning, and reasoning. Meta states it outperforms Google's latest Gemini release in coding and reasoning benchmarks. It also beats older versions of OpenAI and Anthropic's models in some areas. Meta's CEO, Mark Zuckerberg, said the focus is on delivering strong, versatile models at a low cost. While Muse Spark 1.1 is impressive, it still lags behind the flagship models from Anthropic and OpenAI in one coding metric. This release follows Meta's significant entry into the AI race three months ago, under a reorganized unit led by Alexandr Wang. The company spent $14.3 billion in 2025 to acquire a stake in Wang's AI data startup, Scale AI. Muse Spark now powers Meta's AI assistant, and developers can access it through an API. Meta also released Muse Image and Muse Video this week, focusing on image and video generation. These new visual AI models produced impressive results. However, Meta's decision to allow users to apply AI effects to other people's Instagram photos without permission caused controversy. This rapid development shows Meta is accelerating its efforts to compete in the AI space, which could impact how we interact with technology.