The Frontier Design Benchmark is live.

Read on XLinkedIn

Leaderboard

Updated Nov 2026 · 27 Contra jobs
Contra Labs

Frontier
Design Benchmark

Elo rating from blind head-to-head votes

  1. Gestalt
  2. Human golden
  3. Claude Fable 5.2
  4. Claude Opus 5.5
  5. GPT-6.1 Sol
  6. Claude Sonnet 5.5
  7. Kimi K3.1 Pro
  8. GPT-6 Astra
  9. Qwen3.8 Max
  10. Kimi K3
  11. GLM-5.3 Prime
  12. Gemini 4 Carbon
  13. Muse Spark 1.3
  14. Muse Wonder
  15. Grok 4.7
  16. Gemini 4 Argon
  17. DeepSeek V4 Pro

contralabs.com/frontierdesign

#ModelWin rate95% CIVotes
  1. 01 GestaltGestalt · Elo 1931 90% 86–92 420
  2. 02 Human goldenContra designers · Elo 1739 74% 71–78 571
  3. 03 Claude Fable 5.2Anthropic · Elo 1713 73% 69–76 683
  4. 04 Claude Opus 5.5Anthropic · Elo 1691 73% 69–76 527
  5. 05 GPT-6.1 SolOpenAI · Elo 1668 70% 66–74 549
  6. 06 Claude Sonnet 5.5Anthropic · Elo 1667 69% 65–73 440
  7. 07 Kimi K3.1 ProMoonshot AI · Elo 1663 71% 66–74 519
  8. 08 GPT-6 AstraOpenAI · Elo 1638 66% 61–70 449
  9. 09 Qwen3.8 MaxAlibaba · Elo 1434 39% 35–44 428
  10. 10 Kimi K3Moonshot AI · Elo 1415 40% 35–44 425
  11. 11 GLM-5.3 PrimeZ.ai · Elo 1403 41% 37–45 464
  12. 12 Gemini 4 CarbonGoogle · Elo 1366 36% 32–40 546
  13. 13 Muse Spark 1.3Meta · Elo 1335 31% 27–35 503
  14. 14 Muse WonderMeta · Elo 1249 22% 18–25 486
  15. 15 Grok 4.7xAI · Elo 1247 23% 19–26 559
  16. 16 Gemini 4 ArgonGoogle · Elo 1221 20% 17–24 547
  17. 17 DeepSeek V4 ProDeepSeek · Elo 1121 12% 9–15 506
0%25%50% · even75%100%

Ranked by Elo from Contra’s designer panel: 4,311 blind votes by 35 professional designers on 33 briefs. The human golden is ranked with the models.

Pareto frontier

11001200130014001500160017001800 $0.5$1$2$5$10$20$50$100 Mean cost of one output (US$, log scale) Human golden · Elo 1739

How the ranking works

Methodology

Real jobs

Each job is a real client brief from Contra’s marketplace with its files, the client renamed. Every model on the round builds a working website for it, and its golden, a design a person picked, is set beside the outputs.

Blind review

Reviewers see two outputs for the same brief side by side and pick the better one, without seeing which model made which. The sides are shuffled for every pair.

Rubric

We curated 35 design rubrics from widely acknowledged design principles. They catch common mistakes, such as too many fonts or inconsistent spacing, so a score says what to fix. Across the models, rubric scores track the designers’ ranking (r = 0.94).

Your vote ranks the models

Two model outputs for the same brief, live and side by side, without the models’ names. Pick the better one, or call a tie.

Vote on outputs
Sign inGet in touch