Understand which models win on Meta so you can choose the best model for your ad creative. CreativeBench runs independent benchmarks with real Meta spend and live traffic, then publishes public ELO leaderboards for copy, image, video, and voice.
Highlights v2026-08
CreativeBench ELO
Overall ELO from live Meta head-to-heads · Higher is better
$900 in Meta spend · 302,410 impressions · 5 briefs · DTC skincare
Rank ↑
Model
CreativeBench ELO
LPV Rate
Link CTR
Sample
1
Claude Fable 5
1548
1.42%
1.80%
151.2k
2 ▲1
GPT-5.6 Sol
1521
1.31%
1.66%
149.9k
3 ▼1
Gemini 3.1 Pro
1512
1.28%
1.63%
150.6k
4
Grok 4.5
1504
1.22%
1.55%
100.4k
5
Kimi K3
1476
1.11%
1.41%
150.1k
6
DeepSeek V4 Pro
1418
0.91%
1.16%
150.9k
Ranked by ELO from head-to-head wins on live Meta traffic. See methodology →
See the difference for yourself.
Same brief, same budget, same audience. These are this round's top three ads exactly as they ran, ranked by how many people they actually pulled to the landing page.
#1Claude Fable 5LPV 1.87%
S
Saltroot
Sponsored
···
Sunscreen you'll actually want to reapply. Saltroot melts in like a serum. No white cast, no sting, reef-safe SPF 50.
Ad creative placeholder
SALTROOT.CO
SPF that feels like skincare
Free shipping over $30
Shop Now
LikeCommentShare
#2GPT-5.6 SolLPV 1.64%
S
Saltroot
Sponsored
···
Your dermatologist was right about daily SPF. She just never mentioned it could feel this good.
Ad creative placeholder
SALTROOT.CO
The no-white-cast mineral SPF
30-day money-back guarantee
Shop Now
LikeCommentShare
#3Gemini 3.1 ProLPV 1.55%
S
Saltroot
Sponsored
···
4,812 five-star reviews and one stubborn myth: that mineral sunscreen has to feel chalky. It doesn't.
Ad creative placeholder
SALTROOT.CO
Saltroot SPF 50, try it risk-free
Ships free, always
Shop Now
LikeCommentShare
Value frontier
v2026-08
CreativeBench ELO against cost per 1M output tokens. The dashed line connects Pareto-optimal models. Nothing above-left of it exists yet.
How it works
Each CreativeBench round is an independent evaluation: same briefs, equal Meta budgets, live traffic, then a public ELO leaderboard. Methodology details below.
01
Generate
Every model gets the same five ad briefs via API: same product, same audience, same constraints. No human editing.
02
Launch on Meta
10 ads per model, equal budgets, identical audiences and placements. Creation, launch and results collection are fully automated through Admanage.ai. No human touches an ad.
03
Measure
Landing Page View rate over a 5-day flight and ~300,000 impressions of live traffic across PH, ZA, KE and NG.
04
Rank
Bradley-Terry ELO from head-to-head, brief-level wins. Bootstrap confidence intervals on every score.
CreativeBench is an independent AI creative benchmark built by the team at Admanage.ai. We run these evaluations to know which models to build with, and we publish the leaderboards so you can choose models with the same evidence.