LithoBench - Comparative Analysis (Open-Ended): leaderboard

LithoBench Comparative Analysis (Open-Ended) scores vision-language models on 1,200 open-ended questions about discrimination between visually similar lithologies in GF-2 satellite image patches (LithoBench, 2026). The OPEN_J score, from 0 to 10 and higher is better, blends ROUGE-L and BLEU-4 overlap with the reference answer and a GPT-5.4 judge rating.

Metric: OPEN_J score (0-10) = 0.4 x ROUGE-L x 10 + 0.2 x BLEU-4 x 10 + 0.4 x the GPT-5.4 judge's overall score (0-10), comparing the answer with the reference answer, on the 1,200 open-ended comparative analysis questions of LithoBench (discrimination between visually similar lithologies in GF-2 remote-sensing image patches); temperature 0, 1,024 output tokens; higher is better. Source: arxiv.org. Saturation forecast: Around 2033. 11 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3.5 397B A17B3.87#264
2InternVL3.5-8B-HF3.79
3Kimi K2.53.79#266
4GLM-4.6V3.75#710
5Intern-S1-Pro3.73#585
6InternVL3.5-241B-A28B3.68#733
7Grok 43.58#319
8Gemini 3 Pro3.58#145
9Qwen 3.5 9B3.31#828
10Qwen 3.5 27B2.43#375
11Claude Sonnet 4.61.23#173

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

Interactive version: theaggregate.ai/benchmark?slug=lithobench-comparative-analysis-open-ended · How It Works · Data refreshed daily, snapshot 2026-11-21.