Meta Andromeda 2.0 Update: What Changed and What To Do
Meta rebuilt the system that decides which ads even get a chance to show. Here is what the update really says, what it does not say, and what to check in your own ad account.
Quick answer: “Andromeda 2.0” is a nickname used by marketers. Meta did not launch a product with that name. It refers to the next-generation Andromeda ad retrieval system that Meta described in an engineering update on 28 October 2026.
What is Andromeda?
Before Meta shows an ad, it must first pick a small group of candidate ads from a huge pool. Andromeda is the part of Meta’s system that does this first step, called ad retrieval. Later systems then rank those candidates and choose what the person actually sees.
What changed in the October 2026 update
According to Meta’s own announcement, the rebuilt system can look at far more ads in the same time. Here are the main points:
| What Meta says | In plain words |
|---|---|
| New mixture-of-experts design with specialised user models | Different parts of the system handle different kinds of people and situations, so matching is more precise. |
| Over 1 billion ads scored per second in production, about 100 times the 2024 system | Many more ads can be considered for each person, not only a small shortlist. |
| About 300 times fewer computing steps per model check | The system is more efficient, so it can handle more ads at lower cost. |
| Works alongside Meta’s GEM and GenRecs systems | Retrieval, ranking and recommendations are tied together more closely. |
Important: these numbers are Meta’s own measurements. They are not independent benchmarks of what your campaigns will earn.
Why Meta did this
Meta says AI creative tools have helped advertisers produce a huge number of new ads, so the pool of ads to choose from keeps growing. The system needed to handle that scale and also give new ads with little history a fair chance.
What it means for your ads
- Creative does more of the targeting. Meta reads your ad and decides who may like it. Ads with clearly different hooks, angles and formats give the system more to work with than five copies of the same idea.
- Keep campaign structure simple. Many industry teams report better results from fewer ad sets with more creatives inside each, because splitting budget too thin slows learning.
- Treat audiences as suggestions. Broad targeting plus strong creative is now a common starting point. Lookalikes help less than before, according to agency reports.
- Fix your signal quality. Make sure the Meta Pixel and Conversions API send purchases correctly, so the system learns from real buyers.
- Judge by profit, not by hype. Check ROAS against your break-even ROAS, along with CTR, CPC, conversion rate and cost per purchase.
Myths to avoid
- “Andromeda 2.0 is an official Meta product.” It is a nickname for the next-generation system.
- “Meta’s improvement numbers will happen in my account.” They are company measurements across Meta’s platform, not promises for any one brand.
- “I must rebuild my whole account now.” No. Review your evidence and test changes one at a time.
Check this week
- How many truly different creatives (different hook, angle or format) do you have live?
- Are several ad sets splitting a small budget? Can you merge them?
- Is your Pixel and Conversions API sending purchase events correctly?
- Do you know your break-even ROAS and your current cost per purchase?
- Are you testing new creatives every week, with a clear way to decide winners?
Want a simple system for all this?
The Meta 5X ROAS ebook shows you what budget to start with, which numbers to check, how to test campaigns and creatives, and when to scale.
See the ebookBased on Meta’s engineering announcement of 28 October 2026 and industry coverage of it. Results vary by brand, product, offer and budget, and are not guaranteed. Always check Meta’s official sources for the latest changes.