Meta Andromeda 2.0 Explained: What Changed, and What You Should Do Now

Quick answer: Andromeda is Meta's AI system that shortlists which ads each person could see before the auction. Its next generation, nicknamed "Andromeda 2.0" (not an official Meta name), was detailed on 28 October 2026: it now scores every eligible ad with specialised models. To benefit, run broad targeting, feed it genuinely different creatives, and send real lead-quality signals through the Conversions API.

Before your ad ever enters an auction on Facebook or Instagram, it has to pass a silent test. Out of a huge pile of eligible ads, Meta's AI picks a shortlist for each person. If your ad is not on that shortlist, your bid, your budget and your beautiful landing page never get a chance.

That shortlisting system is called Meta Andromeda. On 28 October 2026, Meta published details of its next generation. Marketers quickly started calling it "Andromeda 2.0".

In this guide, I will explain what actually changed, what Meta has and has not said, and the practical steps small businesses and performance marketers should take now. Every number here comes from Meta's own posts, and I have listed the sources at the end.

First, a quick note on the name "Andromeda 2.0"

Let me be honest up front. "Andromeda 2.0" is not an official Meta name. Meta's engineering post calls it "the next generation of Meta Andromeda", "the latest iteration" and "the 2026 Meta Andromeda". The "2.0" label comes from marketers and trade press. I use it in this article because that is what people search for, but the real product name is still simply Meta Andromeda.

What is Meta Andromeda?

Meta shows ads in stages. Think of it like a hiring process.

1. Retrieval (Andromeda): narrows a vast pool of eligible ads down to a small set of strong candidates, in milliseconds.
2. Ranking: heavier models score those candidates in fine detail.
3. Auction: the winning ad is shown.

Meta first described Andromeda on its engineering blog on 2 December 2024. It said retrieval selects "from tens of millions of ad candidates into a few thousand relevant ad candidates". The 2024 system ran on NVIDIA Grace Hopper Superchips and Meta's own MTIA chips, and Meta said it allowed a 10,000x increase in model capacity. At launch, Meta reported a +6% recall improvement and a +8% ads quality improvement on selected segments.

In simple words, Andromeda is the gatekeeper. It decides which of your ads are even considered for a particular person.

What changed in "Andromeda 2.0"?

According to Meta's 28 October 2026 engineering post and its Meta for Business summary published the same day, here are the big changes.

1. It now scores every eligible ad

Meta says the new system, in production, "scores every eligible ad" and surfaces the top candidates at over one billion ads per second. Meta calls this a 100x lift compared to the 2024 system.

Meta's own analogy is helpful. Before, the "concierge" checked the most popular shelves in the store. Now, it "walks the entire store" and reads every product label, in the same amount of time.

2. Specialised "expert" models

The old approach used one general model. The new one uses a mixture-of-experts design, an idea borrowed from large language models. Meta replaced a single user model with an ensemble of specialised user "experts". A routing layer then sends each ad to the most relevant experts.

Meta says this brings "near ranking-level" judgment into retrieval at roughly 300x fewer computing operations than the 2024 retrieval model.

3. A faster, leaner machine

Meta fused the whole retrieval process onto a single GPU server. That includes index matching, which Meta describes as "enforcing advertisers' targeting rules". This cut network traffic by about 100x. Meta also says it worked with NVIDIA from early 2026 to push the hardware further.

4. Real results, as reported by Meta

• Since 2025, Andromeda has delivered a 2.4% increase in conversions on Facebook.
• In Q3 2026, Meta increased Andromeda's complexity by 10x without increasing computing costs, which lifted conversions by 1.37% on Facebook and 0.24% on Instagram Reels.

A small caution: these are Meta's own figures, averaged across advertisers. Your account may see more, less or nothing noticeable.

5. It works with Meta's other AI upgrades

On the same day, Meta explained that GenRecs, its LLM-based recommender, now enriches Andromeda. Large language models read each ad's text and images and describe it in "semantic tokens", such as brand, colour, fit and style. Meta says this helps new ads, refreshed creative and new products get on "a faster track to performing". In Q3 2026, generative retrieval on Facebook Feed increased conversions by nearly 2% across targeted segments.

Andromeda (original) vs Andromeda 2.0: side by side

In plain English: the job did not change, the scale did. The old Andromeda was a very smart filter that looked at part of the pool. The new one looks at the whole pool, judges each ad more like the ranking stage would, and does it faster and cheaper. Your targeting still sets the boundaries. What Meta has not published is any new rule about how many ads to run, how to structure accounts, or how budgets should change. Anyone quoting such rules is giving you their opinion, not Meta's.

Why did Meta do it?

Meta gives a clear reason. Generative AI creative tools "have helped advertisers rapidly create hundreds of millions of new ads", growing the candidate pool to "an unprecedented scale". The 2024 system could not score that many ads with a more accurate model at the needed speed.

Meta also says small gains in retrieval "can drive large gains in advertiser value". And it is upfront about its own benefit: more business from advertisers and lower costs, since each GPU now evaluates more ads per second.

How it affects targeting, creative and budgets

Targeting

Your targeting rules still decide which ads are eligible for which person. Andromeda then picks the best from that eligible pool. So narrow targeting does not "break" Andromeda. It just gives it a smaller pool to choose from. Given that the system can now consider far more ads per person, a broader eligible pool gives it more room to find the right match. That last point is my practical reading, not a Meta statement.

Creative

This is where the update matters most. Meta says more of your ads and creative variations are now considered for each person. But there is a catch.

In December 2025, Meta explained that its system "groups together ads that share key visual and thematic attributes". When ads "look or feel alike, these are seen as variations of the same creative". Ten colour swaps of one video are, in effect, one idea. Distinct ideas give the system genuinely different options to match with different people.

Budgets

Meta has not published budget guidance tied to this update. What Meta did say, in its 28 October 2026 auction explainer, is that "the highest bid or the biggest budget doesn't automatically win". Ad quality reflects the whole customer experience, including what happens after the click. Better matching rewards better ads and better offers, not bigger spending alone.

What to do now: a practical checklist

1. Build creative diversity, not creative volume. Meta's guidance separates iteration (same visual, new CTA) from diversification (different personas, use cases, formats, messages and themes). Make each new ad a new idea. For example, a customer testimonial, a founder explainer, a price-led static and a problem-solution Reel.
2. Keep targeting broad where you can. Let your eligible pool be large and let creative do the targeting. Use exclusions and hard limits only when the business truly needs them.
3. Use Advantage+ where it fits. Meta made Advantage+ leads campaigns available globally in October 2025 and reported 14% lower cost per lead and 10% lower cost per qualified lead on average.
4. Refresh creative regularly. GenRecs is designed to help new ads and products get going faster, so fresh, clearly different creative has a better runway.
5. Make your product and message obvious in the ad. Meta's LLMs read ad text and images to understand what an ad is about. Clear visuals and on-ad copy help that understanding.
6. Feed quality conversion signals. This one deserves its own section.

Feed Meta the right signals with the Conversions API

Andromeda finds the people most likely to act. But "act" means whatever you tell Meta to optimise for. If you only send raw form fills, you will get more form fills, including junk.

Meta's auction explainer says you control "the outcome you optimize for, what each conversion is worth, and how results are measured". For lead generation, that means sending offline lead quality back to Meta:

• Use the Conversions API for CRM to send down-funnel events like "qualified", "site visit" or "sale" from your CRM.
• Pair it with the Conversion Leads performance goal, which works with Instant Forms.
• Meta's developer docs say the conversion event should happen within 28 days of the lead, with a conversion rate between 1% and 40%, and expect a 2 to 4 week learning phase.
• Meta reports that advertisers using Conversions API for CRM with the conversion leads goal see, on average, 19% lower cost per quality lead. In October 2025 it also removed the minimum monthly lead volume requirement.

If your team works leads in spreadsheets or a CRM and has no developer to wire this up, a tool like LeadSmart, a verified Meta Tech Provider that syncs lead-ad leads and sends lead-quality updates back through the Conversions API, can handle it. Whatever tool you use, the principle is the same: tell Meta which leads were good.

Common mistakes to avoid

• Treating 28 October as a switch-on date. Meta describes the system as already in production, and the Q3 gains came before the post. Do not read a before-and-after line into your reports on that date.
• Uploading near-duplicates and calling it testing. Same video, new caption, is iteration, not diversity.
• Over-splitting ad sets. Many tiny ad sets with narrow audiences give each one less data. This is general practice, not a new Andromeda rule.
• Optimising for cheap leads only. Without quality feedback, the system learns to find more of the leads you do not want.
• Believing magic numbers. Meta has not published a required number of creatives or a similarity threshold.
• Expecting the auction to be won by budget. Meta says it is not.

FAQ

Conclusion

"Andromeda 2.0" is not a new button or a new campaign type. It is a much bigger, faster and smarter shortlisting engine that now looks at every eligible ad. The winners will be advertisers who give it real choices and honest feedback: genuinely different creative, room to find people, and clear signals about which conversions and leads actually matter.

Get those three things right, and you are working with Meta's AI instead of against it.

Jaideep Chauhan is the founder of Ndimensions (nds.studio) and LeadSmart (leadsmart.pro), an Indian SaaS that syncs Meta lead-ad leads and sends lead-quality signals back to Meta through the Conversions API.

Sources

1. Engineering at Meta, "Meta Andromeda: Supercharging Advantage+ automation with the next-gen personalized ads retrieval engine", 2 December 2024. https://engineering.fb.com/2024/12/02/production-engineering/meta-andromeda-advantage-automation-next-gen-personalized-ads-retrieval-engine/
2. Engineering at Meta, "Meta Andromeda: Powering the Next Generation of Ads Retrieval Through Model-System Co-Design", 28 October 2026. https://engineering.fb.com/2026/10/28/ml-applications/meta-andromeda-next-gen-ads-retrieval-gpu-co-design/
3. Meta for Business, "Inside the Three New AI Innovations in Ads Ranking Improving your Ad Performance", 28 October 2026. https://www.facebook.com/business/news/three-new-ai-innovations-in-ads-ranking
4. Meta for Business, "How Meta's Ads System Matches Your Ad to the Right Person at the Right Time", 28 October 2026. https://www.facebook.com/business/news/how-metas-ads-system-matches-your-ad-to-the-right-person
5. Meta for Business, "Demystifying Creative Diversification", 16 December 2025. https://www.facebook.com/business/news/demystifying-creative-diversification
6. Meta for Business, "More quality leads this Q5 with Meta's latest AI-enabled updates", 30 October 2025. https://www.facebook.com/business/news/more-quality-leads-this-q5-with-metas-latest-ai-enabled-updates
7. Meta for Developers, "Conversions API for CRM integration" and "Follow-up steps after CRM integration" (accessed 4 November 2026). https://developers.facebook.com/docs/marketing-api/conversions-api/conversion-leads-integration/ and https://developers.facebook.com/docs/marketing-api/conversions-api/conversion-leads-integration/crm-integration/6-follow-up-steps/
8. Segwise, "Meta Andromeda 2.0: What Changed and What to Do Now" (on the unofficial "2.0" label and the 28 October timing caveat), accessed 4 November 2026. https://segwise.ai/blog/meta-andromeda-2-0
9. AdShift, "Meta explains its 2026 ads ranking upgrades", 28 October 2026 (updated 30 October 2026). https://adshift.co/changes/meta-ads-ranking-andromeda-gem-genrecs-2026/