Muse Spark1.1:Meta aims to undercut OpenAI and Anthropic with aggressive pricing

Meta has introduced Muse Spark 1.1, a new AI model, which is now available for free in the Thinking mode of the Meta AI app and in the browser. At the same time, the company is launching a public preview of its new Meta Model API, through which developers and enterprise customers can access Muse Spark 1.1.

For the first time, Meta is relying on a paid interface, supplementing its previous approach, characterized by freely available Llama models, with a commercial offering. With low prices, the company aims to take market share from established AI providers OpenAI, Anthropic, and Google: According to Meta, using the API costs only about a quarter of what OpenAI and Anthropic charge for their top models. However, whether Muse Spark 1.1 delivers comparable performance remains to be seen.

Compared to Muse Spark released in April, version 1.1 is expected to improve primarily in agentic tasks. According to Meta, the model can process complex projects faster by gathering context, creating a plan, and delegating execution in parallel to multiple sub-agents. Muse Spark 1.1 is said to actively manage its context window of one million tokens, retrieve previous work steps, and condense important intermediate steps for later tasks. Meta also mentions improvements in computer control, programming tasks, and multimodal understanding.

Meta also sees progress in security: Muse Spark 1.1 is said to be more resistant to jailbreaks and prompt injection attacks, hallucinate less often, and be less prone to uncritical agreement.

Meta seeks to catch up with AI leaders

Despite these advances, CEO Mark Zuckerberg admitted in a conversation with Bloomberg that Meta's AI models are still partly lagging the competition. A future model codenamed “Watermelon” is expected to help bring Meta closer to the technological forefront.

The Muse models are developed by Meta Superintelligence Labs, Meta's newly established frontier AI unit. With this lab, Meta aims to get closer to the leading AI providers again after the disappointing performance of Llama 4. This week, the unit also introduced Muse Image, Meta's first image generation model from the new lab.

Meta is currently investing heavily in AI data centers and chips, expecting investments of up to 145 billion US dollars in 2026 alone. With the costly AI expansion, the pressure is growing to turn the enormous expenses into a viable business. In addition to AI subscriptions for Meta AI, which are being tested in some countries, and planned AI agents for businesses, the paid Meta Model API is now also intended to contribute to monetization. According to reports, Meta is also considering renting out excess capacity from its AI data centers to third parties.

(rie)