Meta Launches Muse Ember: Its Smallest Model Sits on the Efficiency Frontier
Simon Willison · rss · 2026-11-10
Simon Willison reviews Meta AI's new Muse Ember, its smallest, most efficient model targeting the Claude Haiku / GPT Luna tier, built for agent busywork: tool calls, coding subtasks, routing, extraction, and long document parsing.
Pricing:
- Same as GPT-6 Luna and Claude Haiku 5.5 for short prompts: $0.10/M input, $0.50/M output
- Diverges at long context: Ember jumps to $0.30/$1.50 past 200k tokens, between Luna ($0.20/$0.75 past 272k) and Haiku ($0.50/$2.50 past 100k); Ember has the cheapest cached input ($0.004/M)
- Same-day release Claude Fable 5.2 costs 80x more ($8/$40)
Key assessment:
- Per-token price comparisons matter less now since reasoning token counts differ wildly; the Artificial Analysis Intelligence Index vs. Cost per Task chart is more informative
- Muse Ember sits right on the Pareto line: no model on the full leaderboard scores higher for less per task
- Willison rendered six pelicans on bicycles across Ember's six reasoning levels (minimal to max): "high" failed to return a complete SVG, and reasoning level didn't cleanly map to visual quality; max-level output rivals Muse Wonder at 40x the price
He also notes the grey dot on the Pareto line is GPT-6.1 Terra (low), a family worth more attention.
More from Models
- TuF Labs tops ARC-AGI-2 leaderboard: general approaches now beat domain-specific tricks — rbhar90 · 2026-11-10
- Gemini users hit new science guardrails: model refuses physics chats and suggests therapy — ElderMillennialBrain · 2026-11-10
- Users report GPT acting 'dumber':日常 tasks degrade noticeably — lxfater · 2026-11-10
- Google's Logan Kilpatrick teases with a one-word 'Gemini' reply, hinting at upcoming launch — ammaar · 2026-11-10
- Kimi K3.1 Pro lands on Microsoft Foundry with working agent runs and tracing — lee_stott · 2026-11-10
- SIRIS benchmark: LLMs can't reliably tell when simulator errors invalidate design recommendations — MicrosoftResearch · 2026-11-10