Gamer - Minigame - Janus Maze: leaderboard

40 reduced benchmark task instances; 3 evaluation repetitions. Each task averages VALID attempts; malformed/missing attempts are not treated as zero. Acc is unweighted mean over seven game types, not pooled trial accuracy. Underlying APIs retained; trained Gamer variants and ambiguous Qwen Base withheld. Metric: Task success (%). Source: https://arxiv.org/html/2611.18397v1

Metric: Task success (%). Source: arxiv.org. Saturation forecast: Not forecast. 7 models tracked. Updated 17 Nov 2026.

Top models

#ModelScoreOverall rank
1GPT-5.6 Sol14.3#33
2Gemini 3.8 Flash14.3#39
3GPT-5.6 Luna14.3#130
4GPT-6.1 Sol14.3#12
5DeepSeek-V4-Flash-Vision-Exp14.3#129
6Claude Opus 50#23
7GLM-5.3 Flash0#136

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

Interactive version: theaggregate.ai/benchmark?slug=gamer-minigame-janus-maze · How It Works · Data refreshed daily, snapshot 2026-11-17.