PlantXpert - Visual Challenges: leaderboard

PlantXpert (2026) is an evidence-grounded multimodal benchmark for soybean and cotton phenotyping: four-option questions built on figures, captions and text from plant science publications, evaluated on the expert-annotated Gold split with chain-of-thought prompting. This board is accuracy in percent on the questions whose main challenge is visual recognition; higher is better.

Metric: Accuracy (%) on the visual recognition questions on the expert-annotated PlantXpert Gold split (369 four-option questions on soybean and cotton figures from plant science publications), base checkpoints with chain-of-thought prompting, temperature 0, mean of three runs; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3 VL 8B78.68#917
2Qwen 3 VL 2B77.16#2134
3Qwen 3 VL 30B A3B76.65#784
4Gemma 3 27B75.63#1663
5Qwen 3 VL 4B75.63#1151
6Gemma 3 12B73.6#2059
7InternVL3.5-2B72.59#3284
8LLaVA-NeXT-34B65.99#3443
9Granite-Vision-3.3-2B60.41#6934
10Gemma 3 4B59.9#5339
11LLaVA-Next-LLaMA3-8B54.31#5654

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

Interactive version: theaggregate.ai/benchmark?slug=plantxpert-visual-challenges · How It Works · Data refreshed daily, snapshot 2026-11-21.