Meta's Muse Ember matches Haiku benchmarks at roughly a third of the cost
shoumikgoswami · reddit · 2026-11-10
Meta's Muse Ember (released Nov 9) targets repetitive agent work. Artificial Analysis scores it and Claude Haiku 5.5 the same (36 Intelligence Index), but estimates $0.14 vs $0.41 per benchmark task despite identical headline token prices: Ember's cached input is 60% cheaper and it emitted roughly a third as many output tokens. Caveats: the advertised 1M context's $0.10/$0.50 rates only apply up to 200K prompt tokens, rising to $0.30/$1.50 beyond; and the $0.14 reflects the benchmark workload, not your invoice.
More from Models
- Hugging Face open-sources 14.8B DNA model Carbon-X, second only to GPT-6.1 Sol on GenomeQA — LoubnaBenAllal1 · 2026-11-10
- A bare "你好" in Chinese triggered Claude's frontier-LLM safeguards and paused the session — menhguin · 2026-11-10
- Critic: openly evil AI alignment would be safer than Anthropic's contradictory Claude post-training — ludwigABAP · 2026-11-10
- Dev laments Fable 5.2 demos all reduce to "do something cool" prompt passivity — andimarafioti · 2026-11-10
- Fable-5.2 ranks #4 on EyeBench, ~1.5x more token-efficient than Opus-5.5 — adonis_singh · 2026-11-10
- Fable 5.2 hits 41.3% on Remote Labor Index, nearly double Opus 5.5 — Sea_Emergency_8458 · 2026-11-10