PlantMarkerBench - Evidence Type (Maize): leaderboard

Evidence-type classification on the balanced Maize pilot split of PlantMarkerBench (2026), which asks whether a sentence from a plant biology paper supports a gene as a cell-type marker and what kind of evidence it gives: macro-F1 over expression, localization, function, indirect and noise labels, times 100. Open-weight models are served by Ollama; higher is better.

Metric: Evidence-type macro-F1 (%): macro-averaged F1 over the five evidence labels (expression, localization, function, indirect, noise) on the balanced 600-sentence Maize pilot split, times 100; open-weight models served by Ollama with the default prompt, OpenAI models with the direct prompt, temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 2.5 32B Instruct (Ollama)44.3
2GPT-OSS-20B (Ollama)41.4
3GPT-5.4 Mini40#406
4Qwen 2.5 3B Instruct (Ollama)34.8
5Llama 3.2 3B (Ollama)34.6
6Llama 3.1 8B (Ollama)28.5
7Mistral 7B Instruct (Ollama)26.8
8DeepSeek-R1-8B (Ollama)24.5
9Phi3-14B (Ollama)23
10Qwen 2.5 14B Instruct (Ollama)23
11Qwen 2.5 7B Instruct (Ollama)12.2
12DeepSeek-R1-1.5B (Ollama)8.7
13Qwen 2.5 1.5B Instruct (Ollama)7.1
14Phi3-3.8B-Instruct (Ollama)6.1

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

Interactive version: theaggregate.ai/benchmark?slug=plantmarkerbench-evidence-type-maize · How It Works · Data refreshed daily, snapshot 2026-11-21.