Introducing Muse Code and Muse Spark 1.2
Introducing Muse Code and Muse Spark 1.2 Yet more evidence that the most important characteristic of any model these days is long-sequence agentic tool calling. Meta shipped their own coding agent as part of getting that to work! Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1, with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer…
Meta has unveiled Muse Code and Muse Spark 1.2, showcasing their commitment to the latest model trends - long-sequence agentic tool calling. Muse Spark 1.2 brings a coding-focused update to its predecessor, Muse Spark 1.1, with enhanced code generation, debugging, understanding of codebases, and end-to-end developer workflows.
The model underwent significant training compute scaling on coding tasks while expanding the diversity of training environments. To ensure optimal performance, Muse Spark 1.2 was co-trained alongside Muse Code. This training incorporated rejection sampled harness trajectories and recipe optimizations for goals, compaction, and subagents. Additionally, Muse Code's toolset was integrated to maximize harness compatibility.
Muse Spark 1.2 underwent extensive training on long-horizon coding tasks, including whole-repository generation, large end-to-end projects, and auto-research. For instance, the model produced this pelican riding a bicycle SVG, a slight yet notable improvement over its 1.1 counterpart. Pricing-wise, Muse Spark 1.2 is offered in two different model IDs: muse-spark-1.2 at $1.25/million input and $4.25/million output, and muse-spark-1.2-contributor at a heavily discounted price of $0.10/$0.20 - roughly equivalent to the pricing of other models like GPT-5.6 Luna and Gemini 3.1 Flash-Lite.
Written by urgent.news from Simon Willison's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
