Meta's Muse Ember Matches Haiku's 36 Intelligence Score at 1/3 Cost
Meta released Muse Ember (muse-ember-1.0), an AI model for repetitive coding tasks, achieving an Artificial Analysis intelligence index of 36 — equal to Claude Haiku 5.5 — while costing roughly $0.14 per question versus Haiku's $0.411, though Haiku leads on FrontierCode benchmark (34.9% vs 27.2%).
On November 9, Meta released muse-ember-1.0, a model designed to handle the repetitive small steps that coding agents perform many times a day: invoking tools, editing short code snippets, extracting fields, and reading long documents. Because these steps repeat at high volume, Meta positioned Ember's pricing against the Haiku tier.
Artificial Analysis Intelligence Index
On the Artificial Analysis intelligence index (published November 9), the top-tier Ember scores 36 , matching Claude Haiku 5.5 (also 36), while GPT-6 Luna scores 28 . Meta states that Ember reaches Haiku's best score at approximately one-third the price. The accompanying chart plots cost per question on the horizontal axis and intelligence index on the vertical axis. Ember's two data points fall between $0.10 and $0.20 per question, near the left side. Haiku's maximum point is 36.2 at $0.411 per question. Luna remains below 30 points. The chart notes that Ember's cost is calculated at list price.
Pricing Details
Standard input: $0.10 per million tokens
Cached input: $0.004 per million tokens
Output: $0.50 per million tokens
For prompts exceeding 200,000 tokens: input $0.30 , cached input $0.012 , output $1.50 per million tokens
Context window: 1 million tokens
Maximum output per call: 512,000 tokens
The per-question cost index works out to roughly $0.14 for Ember versus $0.411 for Haiku, confirming the claimed threefold price advantage.
Availability
Ember is accessible via Meta's API and Muse Code, and also on Google Cloud and Oracle Cloud. Billing is based on token usage per call.
Benchmark Breakdown
The following table compares the top-tier configurations of Ember, Haiku, and Luna across several benchmarks. Each model uses its best publicly reported run configuration, not a single unified setup.
Overall Intelligence Index: Ember 36, Haiku 36, Luna 28
FrontierCode: Haiku 34.9% , Ember 27.2% , Luna 25.2% — Haiku leads on this code-maintainability metric.
Terminal-Bench 4.0: Ember 40.9% , Haiku 39.2% , Luna 16.4%
AutomationBench (strict single-pass): Ember 37.8% , Luna 20.2% , Haiku 13.9%
Model Lineage and Internal Comparison
Ember is the first model trained from Muse Wonder Full . Meta allocates difficult problems to Wonder and delegates high-volume repetitive steps to Ember. Compared to Meta's internal Spark 1.3 , Ember's intelligence index is 36 versus Spark's 37 , but the cost per index question is $0.14 for Ember against $2.39 for Spark 1.3 — a dramatic cost reduction for near-parity performance.
Source: Meta, November 9 — https://research.meta.ai/blog/muse-ember
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
