Contra Labs released its Frontier Design Benchmark, where an anonymous stealth model codenamed Gestalt topped the leaderboard with a 1931 Elo rating and 90% win rate, beating the human designer reference (1739) as well as Claude Fable 5.2 and GPT-6.1 Sol. The benchmark drew on 33 real commercial design tasks and 4,311 blind votes from 35 professional designers across five rounds; Gestalt also scored near-human on quality rubrics and AI-slop detection, though its team and architecture remain undisclosed.
Contra Labs
Benchmark organization behind the Frontier Design Benchmark, built with BenchFlow on real commercial design tasks.
Curated Intelligence
IT之家 · 智能时代T2·Models
74 pts
Hacker News AIT2·Models
58 pts
Contra Labs and BenchFlow launched the Frontier Design Benchmark, built on real client design briefs and judged blind by 68 professional designers, with a rubric correlating 0.94 with human ratings. Two stealth models appeared on the leaderboard ahead of Fable 5.2, though the page discloses no identities, parameters, or scores.