RC RANDOM CHAOS

Meta's Muse Ember hits sweet spot on AI efficiency frontier

· via Simon Willison

Original source

Introducing Our Smallest, Most Efficient Model: Muse Ember

Simon Willison →

Meta released Muse Ember, a compact model targeting agent workloads—tool calls, code generation, routing, extraction, and document parsing. It matches the $0.10/$0.50 per-million-token pricing of Claude Haiku 5.5 and GPT-6 Luna on shorter inputs, positioning itself between those two on the pricing curve when million-token thresholds apply. Claude Fable 5.2 commands substantially higher prices at $8/$40.

Muse Ember occupies the efficiency frontier in Artificial Analysis benchmarks, meaning no competing model achieves better performance at lower cost. The model offers six reasoning levels—minimal through max—which the author tested via image generation tasks. Despite the reasoning level variation available, visual quality didn’t track cleanly with reasoning tier. The generated outputs proved comparable to significantly pricier Muse Wonder.

Read the full article

Continue reading at Simon Willison →

This is an AI-generated summary. Read the original for the full story.