Muse Spark Changelog Clarifies Model Roles

Muse Spark’s evolution is becoming a specialization story rather than a simple version race.

• –Spark 1.1 pairs coding with computer use, multimodal input, and a 1-million-token context window.
• –Spark 1.2 is co-trained with Muse Code for multi-file refactors, deep debugging, and sustained agent workflows.
• –Spark 1.3 targets production coding agents, using about 20% fewer tool calls and 25% fewer tokens than 1.2.
• –Developers should evaluate models by workflow fit, not just benchmark scores or version number.