Meta Andromeda
- Also called Andromeda
- Updated October 31, 2026
Meta Andromeda is the machine learning system Meta uses for ads retrieval, the first stage of picking an ad for a person. From tens of millions of eligible ads it picks the few thousand worth ranking for that person, and ranking models and the auction then decide what shows. It runs inside Meta's delivery system, so there is nothing to switch on.
How does Meta describe Andromeda?
Meta's engineering post on the 2026 version:
Meta Andromeda is Meta's machine learning system for personalized ads retrieval, the first stage of the ads recommendation pipeline. Retrieval narrows a vast pool of eligible ads down to a small set of strong candidates in milliseconds; a later, more computationally expensive ranking stage then scores those survivors in fine detail.
The 2024 launch post gives the scale: retrieval selects "from tens of millions of ad candidates into a few thousand relevant ad candidates", and then "larger and more sophisticated ranking models predict people and advertiser value to determine the final set of ads to be shown to the person." Meta built it for a growing pool, as Advantage+ automation and generative AI tools add more eligible ads.
Is Andromeda a setting you can turn on?
No. It is a system inside Meta's ad delivery, not a campaign option. Meta's creative diversification post describes it "working behind the scenes to power more complex AI models that can better personalize which ads we show someone." Meta's October 2026 overview of its ads system says the gains from these models "reach your account with no work from you."
Did the 2026 update merge retrieval and ranking?
No. The October 28, 2026 engineering post keeps two stages: retrieval, then "a later, more computationally expensive ranking stage". What changed is the retrieval model. Meta says it now scores "every eligible ad with near ranking-level model performance at retrieval scale", using "approximately 300x fewer inference-time FLOPs than the 2024 Meta Andromeda retrieval model".
What Meta did combine is the hardware. Three retrieval steps that ran on separate machines now run together, in Meta's words to "co-locate index matching, scoring, and top-K selection on the same GPU." Meta's advertiser summary uses "ranking" more loosely, calling retrieval "the first step in the ranking process" that now "brings near ranking-level judgment into the retrieval stage."
What results does Meta attribute to Andromeda?
| When | Meta's words | Where |
|---|---|---|
| 2024 launch | "+6% recall improvement to the retrieval system, delivering +8% ads quality improvement on selected segments" | Engineering post, 2024 |
| Since 2025 | "Since 2025, Meta Andromeda has delivered a 2.4% increase in conversions on Facebook." | Engineering post, 2026 |
| Q3 2026 | After a "10X" increase in model complexity: "These improvements increased conversions by 1.37% on Facebook and 0.24% on Instagram Reels." | Advertiser summary, 2026 |
These are Meta's figures for its whole system, measured by Meta, and the first applies to "selected segments". None of them is a forecast for one account.
What it means for your ads
- Your targeting still applies. The 2026 post lists "index matching (i.e., enforcing advertisers' targeting rules)" as part of the retrieval pipeline.
- Give retrieval different ads to choose from. The diversification post pairs Andromeda with creative diversification, which it defines as "a wide range of ad creatives with different themes, messages, and visuals to cater to diverse audience segments." Near-copies of one image don't meet that definition. With "a wider range of creatives", Meta says, "businesses can give our ads system more options to work with when testing different variations of ads with different audiences at scale."
- Check where any ad count credited to Andromeda comes from. Meta's one published number, and the popular ones it never published, are compared in how many ads Meta says to run.
Common mistakes
- Blaming Andromeda when a new ad gets no spend. Meta documents it as retrieval. How spend splits between the ads in one ad set comes from delivery predictions, explained in why new Meta ads get no spend.
- Treating it as a learning phase rule. Meta's two Andromeda engineering posts never mention the learning phase, and neither sets a budget or ad-count rule.
Sources
- Meta Andromeda: Powering the Next Generation of Ads Retrieval Through Model-System Co-Design — Engineering at Meta, October 28, 2026 (accessed October 31, 2026)
- Meta Andromeda: Supercharging Advantage+ automation with the next-gen personalized ads retrieval engine — Engineering at Meta, December 2, 2024 (accessed October 31, 2026)
- Inside the Three New AI innovations in Ads Ranking Improving your Ad Performance — Meta for Business, October 28, 2026 (accessed October 31, 2026)
- How Meta's Ads System Matches Your Ad to the Right Person at the Right Time — Meta for Business, October 28, 2026 (accessed October 31, 2026)
- The Creative Advantage: Unlocking the Power of Diversification with Meta Andromeda — Meta for Business, April 22, 2025 (accessed October 31, 2026)
Meta publishes new Andromeda figures as the system changes, so check the source before quoting one.
Terrads makes ads that differ in theme, message and visual, the diversified creative Meta pairs with Andromeda.