LithoBench - Identification and Description (MCQ): leaderboard
LithoBench Identification and Description (MCQ) asks vision-language models to name the lithology of a GF-2 satellite image patch among visually similar distractor categories, on 800 multiple-choice questions of LithoBench (2026). The score is accuracy in percent, higher is better.
Metric: Accuracy (%) on the 800 multiple-choice identification-and-description questions of LithoBench: pick the lithology of a GF-2 remote-sensing image patch among visually similar distractor categories (an unparsable answer counts as wrong); temperature 0, 1,024 output tokens; higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 12 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 9B | 92.5 | #828 |
| 2 | Gemini 3 Pro | 77.62 | #145 |
| 3 | GLM-4.6V | 76.25 | #710 |
| 4 | Kimi K2.5 | 75.88 | #266 |
| 5 | Intern-S1-Pro | 70.38 | #585 |
| 6 | Grok 4 | 67 | #319 |
| 7 | GPT-5.4 | 66.5 | #135 |
| 8 | Qwen 3.5 397B A17B | 65.38 | #264 |
| 9 | InternVL3.5-241B-A28B | 62.88 | #733 |
| 10 | InternVL3.5-8B-HF | 58.13 | |
| 11 | Claude Sonnet 4.6 | 56.25 | #173 |
| 12 | Qwen 3.5 27B | 43.13 | #375 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=lithobench-identification-and-description-mcq · How It Works · Data refreshed daily, snapshot 2026-11-21.