Single 2.6B-parameter model achieves state-of-the-art performance across drug discovery benchmarks while running entirely on private pharmaceutical infrastructure
CAMBRIDGE, Mass., March 3, 2026 /PRNewswire/ — Insilico Medicine and Liquid AI today announced a partnership that creates lightweight scientific foundation models for pharmaceutical research. The collaboration has produced LFM2-2.6B-MMAI (v0.2.1), available now – a single checkpoint trained to perform at state-of-the-art levels across multiple drug discovery subdomains, not a patchwork of separate point models.
The partnership tackles a critical challenge facing pharmaceutical companies today: how to harness cutting-edge AI capabilities without sending proprietary molecules, assays, and target data to external cloud services. By combining Liquid AI’s efficient LFM architecture with Insilico’s MMAI Gym, (a comprehensive training platform with over 1,000 pharmaceutical benchmarks), the work shows that on-premise deployment can deliver competitive results across the full spectrum of drug discovery tasks in a single system.
The model covers the complete discovery loop, spanning property prediction and ADMET endpoints, multi-parameter molecular optimization, target-aware scoring with protein-pocket conditioning, functional group reasoning, and retrosynthesis planning. Training involved approximately 120 billion tokens of pharmaceutical data across over two hundred different tasks.
“With LFM2-2.6B-MMAI, we’ve shown that efficient architecture design, not just scale, is what makes foundation models practical for the sciences. A single 2.6B-parameter model now matches or outperforms systems ten times its size across the drug discovery pipeline, all on private infrastructure. Our collaboration with Insilico is proof that you can reduce the cost of intelligence while raising the quality bar,” says Ramin Hasani, CEO and co-founder of Liquid AI.
At just 2.6B parameters, the model achieves cloud-scale performance while operating entirely on private infrastructure:
- Property Prediction: Outperformed TxGemma-27B, a model more than 10x larger, on 13 of 22 tasks covering pharmacokinetics and toxicology, and achieved state-of-the-art results on three of these tasks when compared to specialist models built for individual tasks
- Molecular Optimization: Reached success rates of up to 98.8% on industry-standard multi-parameter optimization benchmarks (MuMO-Instruct)
- Affinity Prediction: On Insilico’s internal benchmark – featuring 2.5M experimental measurements across 689 protein targets – produced better correlation scores than frontier models including GPT-5.1, Claude Opus 4.5, and Grok-4.1
- Chemical Reasoning: Demonstrated strong functional group reasoning capabilities (FGBench) and high-quality single-step retrosynthesis suggestions (ChemCensor metric)
These capabilities unlock immediately useful applications for pharmaceutical companies, particularly in high-frequency ADMET screening, medicinal chemistry-facing lead optimization, and retrosynthesis feasibility assessment that prevents wasted experimental effort.
“We are pleased to collaborate with Liquid AI to develop the next generation of lightweight liquid foundation models capable of performing multiple scientific tasks with state-of-the-art performance across drug discovery benchmarks,” says Alex Zhavoronkov, CEO of Insilico Medicine. “Highly-efficient liquid science models will make it easier for more scientists to achieve their goals in order to compress discovery timelines and ultimately help patients.”
About Liquid AI: Liquid AI builds Liquid Foundation Models (LFMs) based on dynamical systems and signal processing. Founded by researchers from MIT, Liquid AI focuses on AI models that are efficient and can be deployed on-premise or in resource-constrained environments. For more information, visit liquid.ai.
About Insilico Medicine: Insilico Medicine is a clinical-stage biotechnology company using AI for drug development across cancer, fibrosis, immunity, central nervous system diseases, and aging-related conditions. The company’s AI platform covers target discovery, molecular design, and clinical development. For more information, visit insilico.com.
About MMAI Gym for Science: MMAI Gym for Science is a domain-specific training environment designed to elevate general-purpose and frontier Large Language Models (LLMs) into pharmaceutical-grade engines for drug discovery and development. Developed by Insilico Medicine as a core component of its Pharmaceutical Superintelligence (PSI) roadmap, the Gym utilizes specialized tracks for Chemical Superintelligence (CSI) and Biology/Clinical Superintelligence (BSI) to teach models domain-specific reasoning across medicinal chemistry, biology, and clinical planning.
The curriculum leverages high-quality reasoning datasets and multi-task fine-tuning to achieve up to 10x performance gains on mission-critical R&D tasks compared to baseline models. To ensure robust and reliable performance, all models are evaluated against a rigorous suite of proprietary and public benchmarks which are meticulously cleaned to avoid data leakage between training and test sets. MMAI Gym for Science is offered through flexible membership programs tailored to pharma and biotech companies, AI labs, and cloud providers looking to transform generalist AI into robust scientific specialists. For more information or to explore membership options, please contact [email protected].
