Build and orchestrate multi-agent workflows with one model — including computer use, which lets the model operate a computer the way humans do, navigating across apps and interfaces to complete long-horizon tasks.
Advanced coding capabilities
Muse Spark benchmarks
Muse Spark is trained to deliver competitive agentic and coding performance, with multimodal perception built in.
Build with Muse Spark, now available on Meta Model API | AI Developers blogGet started with Muse Spark on Meta Model API: how to make your first call, the coding primitives the model is tuned for, and the agentic patterns that get the most out of it.Read more
Muse Spark writes, reviews, and ships code with fewer steps and lower latency. Whether you're building coding agents, review bots, or using AI as a development partner, it's competitive with the best.
Native multimodal perception
Muse Spark perceives video, images, and documents, and its visual reasoning runs through a real execution environment instead of scripted steps. Feed it a screenshot or a clip and let it build, so you can focus on shipping features fast.
Gemini 3.1 ProGoogle
Opus 4.8Anthropic
GPT 5.5OpenAI
Agents
MCP AtlasScaled tool use
88.1
82.2
78.2
82.2
75.3
JobBenchProfessional tool use
54.7
17.0
15.9
48.4
38.3
Toolathlon-VerifiedPersonal tool use
75.6
49.4
61.1
76.2
73.5
OSWorld-VerifiedAgentic computer use
80.8
53.3
76.2
83.4
78.7
Humanity's Last ExamMultidisciplinary reasoning (w/ tools)
