90% win rate · 1,931 Elo
Leading blind preference
Across 4,311 blind head-to-head votes, Gestalt outperformed human reference designs (+192 Elo) and leading commercial models including Claude Fable 5.2 (+218 Elo) and GPT-6.1 Sol (+263 Elo).
Frontier Design Benchmark · #1
Gestalt is an AI design model built for real-world websites and application interfaces. In the Frontier Design Benchmark by Contra Labs and BenchFlow, it ranked #1 in double-blind evaluations by professional designers.
| Rank | Entry | Elo | Gap | Win rate | 95% CI |
|---|---|---|---|---|---|
| #1 | Gestalt (stealth) | 1,931 | Top | 90% | 86–92% |
| #2 | Human designers | 1,739 | -192 | 74% | 71–78% |
| #3 | Claude Fable 5.2 | 1,713 | -218 | 73% | 69–76% |
| #4 | Claude Opus 5.5 | 1,691 | -240 | 73% | — |
| #5 | GPT-6.1 Sol | 1,668 | -263 | 70% | 66–74% |
| #6 | Claude Sonnet 5.5 | 1,667 | -264 | 69% | 65–73% |
| #7 | Kimi K3.1 Pro | 1,663 | -268 | 71% | 66–74% |
| #8 | GPT-6 Astra | 1,638 | -293 | 66% | 61–70% |
The November 18, 2026 research report recorded Gestalt at 1,928 Elo. The table above reflects the Top 8 live leaderboard as checked on November 21, 2026. Win rate reflects overall blind head-to-head comparisons. Research report
90% win rate · 1,931 Elo
Across 4,311 blind head-to-head votes, Gestalt outperformed human reference designs (+192 Elo) and leading commercial models including Claude Fable 5.2 (+218 Elo) and GPT-6.1 Sol (+263 Elo).
Score 83 · r = 0.94
On 35 criteria covering structure, visual craft, content, and usability, Gestalt (83) matches Claude Fable 5.2 and Opus 5.5 (84) and exceeds the human baseline (74), while leading strongly in whole-page taste.
Slop Detector · 0–100
The benchmark’s Slop Detector measures generic AI layout tells such as repetitive card grids, predictable typography, and awkward spacing. Gestalt sits close to human reference designs at the low end.
Rather than one-line toy prompts, the Frontier Design Benchmark tests 33 commercial briefs—27 adapted from real Contra client commissions and 6 from startup partners—judged double-blind by 35 professional designers (median 8+ years of experience).
In the Frontier Design Benchmark, Gestalt is evaluated on full commercial briefs across four categories: marketing websites, landing pages, web applications, and mobile interfaces, each with brand assets and multi-section requirements.
On the November 21, 2026 leaderboard snapshot, Gestalt leads at 1,931 Elo (90% win rate, 95% CI 86–92%), ahead of human reference designs (1,739 Elo, 74%), Claude Fable 5.2 (1,713 Elo, 73%), Claude Opus 5.5 (1,691 Elo, 73%), and GPT-6.1 Sol (1,668 Elo, 70%).
The Design Critique Rubric checks 35 concrete criteria across structure, visual craft, content, and usability (correlating at r = 0.94 with expert preference). The Slop Detector scores generic AI layout clichés on a 0–100 scale, where Gestalt scores near human baselines.
Contra Labs publishes side-by-side benchmark outputs in its official Design Gallery (contralabs.com/frontierdesign/gallery). Gestalt is currently tested in stealth with no public API yet.