Histoboard - Pathology Foundation Model Benchmarks

THUNDER Benchmark
pathology
H&E

Comprehensive benchmark evaluating pathology foundation models across KNN classification, linear probing, few-shot learning, segmentation, calibration, and adversarial robustness tasks.

22 models evaluated

6 tasks

Organs:

multi-organ

Detailed Results

Model	Rank sum	KNN Classification F1-score	Linear Probing F1-score	Few-shot Learning F1-score	Segmentation Dice	Calibration ECE (%)	Adversarial Attack ASR (%)
1UNI2	18	0.833	0.857	0.798	0.690	3.9%	31.7%
2Virchow2	27	0.829	0.848	0.739	0.693	3.9%	31.1%
3H0-mini	38	0.797	0.838	0.750	0.691	3.8%	34.3%
4Midnight-12k	40	0.799	0.847	0.715	0.688	2.9%	37.0%
5UNI	41	0.808	0.835	0.781	0.678	3.8%	40.3%
6H-optimus-1	47	0.825	0.851	0.773	0.645	3.5%	57.4%
7KEEP	48	0.815	0.832	0.771	0.680	4.0%	44.9%
8H-optimus-0	52	0.814	0.838	0.762	0.652	4.0%	43.9%
9Prov-GigaPath	56	0.795	0.829	0.755	0.635	3.4%	42.1%
10Hibou-L	61	0.786	0.837	0.738	0.686	4.7%	39.5%
11Hibou-B	61	0.789	0.812	0.763	0.678	3.2%	52.7%
12Virchow	64	0.774	0.828	0.718	0.692	4.5%	38.3%
13Kaiko ViT-B/16	67	0.787	0.814	0.764	0.668	5.0%	38.8%
14OpenMidnight	70	0.793	0.847	0.437	0.691	5.4%	38.3%
15Kaiko ViT-S/16	73	0.782	0.817	0.751	0.668	4.6%	42.5%
16CONCH	74	0.788	0.819	0.734	0.683	4.1%	57.3%
17CONCH 1.5	76	0.799	0.824	0.750	0.688	4.6%	75.8%
18Phikon	78	0.757	0.809	0.736	0.680	5.8%	33.5%
19Phikon-v2	84	0.739	0.797	0.718	0.674	3.9%	43.8%
20MUSK	98	0.777	0.811	0.719	0.651	4.0%	71.9%
21PLIP	129	0.702	0.740	0.641	0.585	4.5%	60.8%
22QuiltNet	129	0.704	0.739	0.656	0.589	5.8%	55.6%

THUNDER BenchmarkpathologyH&E

Detailed Results

THUNDER Benchmark
pathology
H&E