EarthVerse: leaderboard
Core percentage on 405 package-scoped scientific investigations, averaging answer correctness and process-rubric score. Model and controller configurations remain distinct, including GPT-5.5 under EarthVerse versus GeoMMAgent.
Metric: Core Score (%, controller-specific). Source: arxiv.org. Saturation forecast: Around January 2027. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 | 84.97 | #16 |
| 2 | GPT-5.6 Sol | 81.68 | #19 |
| 3 | GPT-5.6 Terra | 80.06 | #41 |
| 4 | GLM-5.2 | 79.17 | #73 |
| 5 | GPT-5.5 | 78.06 | #31 |
| 6 | Claude Sonnet 4.6 | 76.16 | #90 |
| 7 | Kimi K2.5 | 75.32 | #144 |
| 8 | DeepSeek V4 Flash | 73.54 | #137 |
| 9 | Hy3 | 72.77 | #93 |
| 10 | Claude Haiku 4.5 | 72.23 | #280 |
| 11 | Qwen 3.7 Plus | 70.35 | #92 |
| 12 | Doubao Seed 2.0 Mini | 62.99 | #184 |
| 13 | Gemini 3.1 Flash Lite | 59.66 | #191 |
| 14 | MiniMax-M2.7 | 44.71 | #258 |
| 15 | GPT-4o | 39.23 | #343 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=earthverse · How It Works · Data refreshed daily, snapshot 2026-10-20.