Overview
- Meta began rolling out Muse Spark 1.3 on Wednesday through Muse Code and the Meta Model API, with a higher 'max reasoning' mode held back for more safety testing.
- Meta says the model uses about 20% fewer tool calls and 25% fewer tokens than version 1.2, which the company says reduces latency and cuts per‑task compute and billing costs.
- Independent benchmarks from Artificial Analysis placed Muse Spark 1.3 near the top of its index (a 62 score), and Meta AI chief Alexandr Wang posted performance charts comparing Muse favorably against some Google Gemini variants.
- Meta reports better long‑horizon agentic behavior and coding performance: the model asks clarifying questions, tracks multi‑step plans, avoids needless turns, and shows improved resistance to prompt injections and other adversarial inputs.
- The release is part of an aggressive push that produced four Muse Spark versions in five months and signals a strategic shift away from primarily open‑weight Llama releases toward monetized, API‑gated products while still offering some open variants such as Muse Glimmer.