Kaggle Plat Learning Benchmark: leaderboard
Community benchmark on Kaggle by dagloxkankwanda: six text-game engines (probabilistic reward, associative, concept, language, social observation and sequence learning). A game scores 1 for a win and half the validated progress for a loss, with random play near 0.05 and harness errors counted as losses; the board is the mean over the six engines as a percentage, higher is better.
Metric: Mean Game Score (%). Source: www.kaggle.com. Saturation forecast: Not forecast. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 86.07 | #64 |
| 2 | Gemma 4 31B | 85.12 | #212 |
| 3 | Gemini 3 Flash (Preview) | 83.52 | #90 |
| 4 | GLM-5 | 77.82 | #149 |
| 5 | Gemini 3.5 Flash | 77.22 | #66 |
| 6 | Claude Opus 4.6 | 75.78 | #67 |
| 7 | Gemini 2.5 Flash | 75.25 | #255 |
| 8 | Claude Opus 4.8 | 74.78 | #39 |
| 9 | Claude Opus 4.7 | 74.55 | #51 |
| 10 | GPT-5.5 | 74.12 | #33 |
| 11 | GPT-5.4 | 73.97 | #83 |
| 12 | DeepSeek V3.1 | 72.32 | #275 |
| 13 | Grok 4.20 (Reasoning) | 68.37 | #179 (Grok 4.20) |
| 14 | Claude Sonnet 4.6 | 64.83 | #94 |
| 15 | GPT-OSS-120B | 57.47 | #346 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=kaggle-plat-learning-benchmark · How It Works · Data refreshed daily, snapshot 2026-11-05.