Muse Spark 1.1 scores 69 on Artificial Analysis Coding Agent Index, nipping at GPT-5.5’s heels

Meta just made the AI coding wars a lot more interesting. Muse Spark 1.1, the company’s upgraded multimodal reasoning model, scored 69 on the Artificial Analysis Coding Agent Index using the Opencode harness, putting it within striking distance of GPT-5.5 and outperforming Claude Opus 4.8.

The model launched on July 9, 2026, from Meta Superintelligence Labs. It arrived just three months after the original Muse Spark debuted on April 8, which is the kind of iteration speed that makes competitors nervous.

What Muse Spark 1.1 actually brings to the table

Muse Spark 1.1 is designed specifically for agent-based applications, with a focus on tool use and computer operation. The model ships with a 1 million token context window, allowing it to hold roughly the equivalent of several full codebases in memory at once, which matters enormously when an AI agent needs to understand how different parts of a project connect before making changes.

It also supports delegating tasks to sub-agents, targeting the increasingly popular “agentic” coding paradigm, where AI doesn’t just answer questions but actively performs multi-step development work including bug diagnosis and feature implementation.

Early adopters already include Replit, Cline, and Box.

The pricing play that changes the math

Muse Spark 1.1 is available through Meta’s first paid developer API at $1.25 per million input tokens and $4.25 per million output tokens. New users get $20 in free credits. Those rates position it as substantially cheaper than comparable offerings from OpenAI and Anthropic, while delivering competitive benchmark results.

What this means for the AI competitive landscape

Meta releasing a paid API is itself the headline beneath the headline. The company has historically positioned its AI models as open-source alternatives, with Llama becoming the backbone of countless startups and enterprise deployments. Muse Spark 1.1 signals a strategic pivot toward capturing revenue from developers who want managed, high-performance AI services without running their own infrastructure.

The partnerships with Replit and Cline are particularly worth watching. Replit has millions of developers on its platform, and if Muse Spark 1.1 becomes a default or preferred backend for its AI coding features, that’s the kind of distribution that compounds quickly. Box, meanwhile, represents the enterprise workflow market, suggesting Meta is targeting both individual developers and large organizations simultaneously.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.