Meta undercuts Claude Haiku 5.5 on cached input with Muse Ember

reported by 3 outlets· 3 articles · confidence: high · first seen 2026-11-09 18:00 UTC

What this means for you

If you route high-volume agent work to Haiku 5.5 or Luna, Ember is worth a bake-off on cached-input cost alone: $0.004 per million tokens against $0.01. Note that above 200k prompt tokens Ember costs more per input token than Luna does below its own 272k threshold — measure your actual prompt lengths before switching.

Meta released Muse Ember, its smallest model, aimed at the repeated work agents do — tool calls, coding subtasks, routing, extraction, parsing long documents — with up to a million tokens of context. On short prompts it costs $0.10 per million input tokens and $0.50 per million output, the same as GPT-6 Luna and Claude Haiku 5.5; TechCrunch calls it significantly cheaper, but the difference is cached input at $0.004 per million against $0.01. Prices rise above 200k prompt tokens. MarkTechPost reports it ties Haiku 5.5 on Artificial Analysis's Intelligence Index at roughly a third of the cost per task; no evaluation date is given.

Models in this story

Key facts

  • ·Muse Ember is priced at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 200,000 tokens source
  • ·Above 200,000 prompt tokens, Muse Ember's prices rise to $0.30 per million input, $0.012 per million cached input and $1.50 per million output source
  • ·Muse Ember's cached input costs $0.004 per million tokens, against $0.01 for GPT-6 Luna and Claude Haiku 5.5 source
  • ·Meta describes Muse Ember as its smallest and most efficient model, built for tool calls, coding subtasks, routing, extraction and long document parsing source
  • ·Muse Ember supports up to one million input tokens and has six reasoning levels source
  • ·MarkTechPost reports Muse Ember ties Claude Haiku 5.5 on the Artificial Analysis Intelligence Index at roughly a third of the cost per task; no evaluation date is given source

What the sources say

  • TechCrunch AI — Frames the launch on price, comparing Ember with Haiku and Luna at ten cents per million input tokens.
  • Simon Willison — Walks the tiered price table prompt-size by prompt-size and reads the Artificial Analysis cost-per-task chart.
  • MarkTechPost — Leads on the Intelligence Index tie with Haiku 5.5 at about a third of the per-task cost.

Sources

The original reporting. Follow these — they did the work.

← the wire