GPT 6.2 takes SOTA on ClutterBench, a leaderboard long dominated by Claude models
sandersted · x · 2026-11-19
A shared ClutterBench leaderboard shows GPT 6.2 surprisingly taking SOTA — the benchmark was previously dominated by Claude models. The leaderboard measures how much of each model's graphic the grader would cut (lower is better), and with a simple "keep it simple" prompt added, top models mostly drop to 0-2%, while Google's models all rank below 12th.
Sol 6.2 (xhigh effort) leads at 16%, followed by Astra 6.2 at 24%, Sonnet 5.5 at 29%, and Opus 5.5 at 31%.
More from Models
- Oxford paper: post-training makes LLM worldviews bleaker, models converge on 'spiritual bliss' — twiecki · 2026-11-19
- Artificial Analysis adds site-wide reasoning-effort filters for model benchmarks — ArtificialAnlys · 2026-11-19
- Artificial Analysis Adds Reasoning-Effort Filters Across All Model Charts — ArtificialAnlys · 2026-11-19
- ApprenticeBench: closed models hit 77% vs 22% for open, widening capability wall — ysu_nlp · 2026-11-19
- Microsoft open-sources Aion 2.3B on-device SLM, shipping as the Windows built-in model — jacek2023 · 2026-11-19
- Local model reads a 131K-token patient history in 61s, then reasons for 11 minutes — maniac518 · 2026-11-19