Meta: Muse Ember

Meta’s Muse Ember demonstrates impressive versatility, functioning as an efficient engine tailored for agentic workflows and complex execution. Across evaluations, the model typically performs in the top tier for speed while offering competitive pricing. It also exhibits exceptional reliability, posting a flawless 100% success rate with zero technical failures across all benchmarks. Performance across categories is consistently strong, with standout results in instruction following (87.0% accuracy, 91st percentile) and coding knowledge (95.0% accuracy, 86th percentile). Muse Ember also excels at multi-step reasoning, posting 96.0% accuracy in the 75th percentile with notably rapid execution. While mathematics remains a capable area at 93.0% accuracy (61st percentile), it represents the model’s relative weakness, requiring longer runtimes and higher processing costs compared to other tasks. Overall, Muse Ember's primary strengths are its elite instruction adherence, robust code generation, and rock-solid dependability. Paired with an expansive 1-million-token context window, it stands out as a cost-effective, high-performing solution for routing, tool invocation, and continuous agent-driven subtasks.