Meta: Muse Spark Open

Meta's Muse Spark Open, an open-weight model optimized for agentic coding workflows, demonstrates a strong overall performance profile. While its speed performance is moderate, ranking in the 37th percentile across benchmarks, its pricing tends to be at premium levels, placing it in the 18th percentile. A standout feature is its exceptional reliability, achieving a perfect 100% success rate across all evaluated benchmarks, indicating consistent and dependable operation. In terms of specific benchmark results, Muse Spark Open excels in Instruction Following, achieving 86.0% accuracy (90th percentile), and shows very strong performance in Coding with 94.0% accuracy (78th percentile). Its Reasoning capabilities are also robust at 92.0% accuracy (64th percentile). Mathematics performance is solid at 91.0% accuracy (48th percentile). The model's key strengths lie in its high accuracy across complex tasks, particularly in instruction following and coding, which aligns with its design for agentic coding workflows. Its deep codebase understanding and reliable tool calling are likely contributors to these strong results. The primary area for potential improvement is its premium pricing, which may impact its accessibility for some users.