Meta's Muse Spark 1.2 Model Integrated into Agent Arena Benchmark
Meta's Muse Spark 1.2 reasoning model has been added to the Agent Arena, allowing users to test its performance in Battle Mode and Agent Mode. The benchmark evaluates the model's capabilities on millions of real-world, long-horizon agentic tasks. This integration allows the AI community to benchmark Meta's latest reasoning model against other leading AI agents in handling complex, multi-step tasks with tool access. It provides transparent, real-world performance data for a model designed specifically for agentic workflows. Muse Spark 1.2 features a 1-million-token context window and supports multimodal inputs including text, images, video, audio, and PDFs. Within the Agent Arena environment, the model can access web search, filesystems, and other tools to complete tasks.
## BACKGROUND
Agent Arena is a dynamic benchmarking platform that ranks AI models based on their ability to orchestrate tools and complete complex, long-horizon tasks. Meta released Muse Spark 1.2 alongside Muse Code, a terminal-based AI coding agent, to target advanced reasoning and developer workflows.