Meta released the Muse Spark 1.1 AI model on Thursday, marking its entry into the competitive agentic coding tool market. The multimodal model is available immediately in Thinking mode on the Meta AI app and website, while developers can access it via the newly launched Meta Model API in public preview for US accounts. As an incentive, Meta is offering $20 in free credits for each new API account.
This update comes three months after the initial Muse Spark release in April, which Meta described as a "step-change" over its predecessor. Version 1.1 handles more advanced coding, can detect and repair complex bugs, and runs agentic workflows across multiple applications, including multi-agent setups. It also features native multimodal perception, enabling it to read images, videos, and documents.
Pricing is set at $1.25 per million input tokens and $4.25 per million output tokens, positioning it slightly above Anthropic’s Claude Haiku 4.5 and OpenAI’s GPT-5.6 Luna but within a competitive range that could pressure rivals as enterprises evaluate cost at scale.
Meta’s AI chief, Alexandr Wang, signaled further ambition, saying a larger model codenamed Watermelon is in training and has caught up with OpenAI’s GPT-5.5 on key benchmarks, according to internal sources. Mark Zuckerberg promoted the launch on X for the first time in three years, calling Spark "a strong agentic and coding model at a very low price" and hinting at more to come.
The push is backed by heavy investment: Meta expects to spend $125–145 billion this year on chips, data centers, and infrastructure, including its Iris AI chip due in September. Meanwhile, its stock has been volatile, recently dropping from a record $796.25 to around $580.50 in pre-market trading.
