WeightBench - Macro SWB (38 tasks, no TR-02 loops): leaderboard

Default difficulty, seed0. Claude2samples/task; DeepSeek/GPT1sample/task. Quarantined/invalid submissions score0 (unreviewed lower bounds). Task scores average across runs then tasks; macro averages5tracks. GPT excludes4 loop Fix tasks; use common38task macro, never pool GPT9taskFix with13taskFix. Metric: Native S_WB (0–1). Source: https://arxiv.org/html/2611.18553v1

Metric: Native S_WB (0–1). Source: arxiv.org. Saturation forecast: Around September 2027. 5 models tracked. Updated 17 Nov 2026.

Top models

#ModelScoreOverall rank
1GPT-6.1 Sol0.568#12
2Claude Opus 5.50.546#7
3Claude Sonnet 5.50.491#21
4Claude Haiku 5.50.428#76
5DeepSeek V4 Flash0.232#265

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=weightbench-macro-swb-38-tasks-no-tr-02-loops · How It Works · Data refreshed daily, snapshot 2026-11-17.