Frontier Design Benchmark - Elo: leaderboard
Models build working websites from 33 client briefs (27 rewritten from real jobs posted on Contra's freelance marketplace, 6 real projects from startup partners with their own brand files), and professional designers on Contra pick the better of two outputs for the same brief without seeing which model made which; a Bradley-Terry model turns the votes into Elo. Unlike crowd-voted design arenas, every voter is a working designer and the briefs are real client work rather than one-line prompts. Designer-made reference sites are rated in the same votes as the human bar and are not listed as a row. Contra Labs and BenchFlow ran the released models through OpenRouter, each in its own lab's agent harness where possible; their write-up sets the stealth entrant, Gestalt, apart from the models they ran.
Metric: Elo (blind designer votes). Source: contralabs.com. Saturation forecast: Around March 2028. 16 models tracked. Updated 19 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gestalt | 1931 | ||
| 2 | Claude Fable 5.2 | 1713 | #1 | |
| 3 | Claude Opus 5.5 | 1691 | #9 | |
| 4 | GPT-6.1 Sol | 1668 | #15 | |
| 5 | Claude Sonnet 5.5 | 1667 | #24 | |
| 6 | Kimi K3.1 Pro | 1663 | #35 | |
| 7 | GPT-6 Astra | Ultra | 1638 | #10 |
| 8 | Qwen 3.8 Max | 1434 | #66 | |
| 9 | Kimi K3 | 1415 | #61 | |
| 10 | GLM 5.3 Prime | 1403 | #81 | |
| 11 | Gemini 4 Carbon | 1366 | #5 | |
| 12 | Muse Spark 1.3 | 1335 | #48 | |
| 13 | Muse Wonder | 1249 | #34 | |
| 14 | Grok 4.7 | 1247 | #51 | |
| 15 | Gemini 4 Argon | 1221 | #18 |
No result here: #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=frontier-design-benchmark-elo · How It Works · Data refreshed daily, snapshot 2026-11-20.