Insilico Science AI Bench | Insilico Medicine

Advancing AI Across the Sciences

Evaluating AI in Science & Discovery

A comprehensive benchmarking platform for AI in science — spanning drug discovery, biology, chemistry, materials science, and beyond. A research initiative by Insilico Medicine.

Hosted Benchmarks

Insilico Science AI Bench Benchmarks Explorer

Interactive exploration of benchmarks across Biology, Chemistry, Materials, and Longevity. Browse and compare model performance with built-in visualizations.

Global Leaderboard

Models Ranked: 8 of 12 total
Benchmarks: 253 of 296 total
Categories: 9 active

Top Model: GPT 5.5
Insilico Score: 67.6

Leaderboard Settings

Scoring Mode: Insilico Score, Relative Insilico Score, Mean Rank

284 benchmarks included in computing the Insilico Score.

Included benchmarks (284)

Aging Prediction: Balanced accuracy[0, 1]
GTEx Expression: accuracy_score[0, 1]
NHANES Mortality Classification: Balanced accuracy[0, 1]
TCGA Survival Prediction: Balanced accuracy[0, 1]
...

Ranked Table

#	Model	Insilico Score	Biology	Affinity and Binding	Chemical Synthesis	ADMET, PK & Safety	Clinical Trials	Biologics	Materials	Longevity & Aging	Molecular Design & Optimization
1	GPT 5.5	67.6	34	71	29	72	81	56	88	66	49
2	Gemini 3.1 Pro	60.3	32	64	29	73	78	54	—	62	50
3	Claude Opus 4.8	59.1	28	58	25	70	78	55	—	61	49
4	Grok 4.5	58.1	29	56	28	66	78	52	—	68	47
5	Claude Opus 4.7	56.4	—	71	16	70	—	56	84	—	33
...	...	...	...	...	...	...	...	...	...	...	...

About Insilico Science AI Bench

Insilico Science AI Bench is a broad benchmarking platform developed by Insilico Medicine to evaluate AI across scientific domains — from drug discovery and biology to materials science and longevity research.

Multi-Domain Science: Benchmarks spanning drug discovery, chemistry, biology, materials science, and longevity.
AI Model Evaluation: Testing large language models and specialized AI systems on diverse scientific tasks.
Transparent Metrics: Clear, reproducible evaluation metrics with public leaderboards.
Research-Backed: All benchmarks supported by peer-reviewed scientific publications.