Insilico Medicine Enhances MMAI Gym with Benchmark Leaderboard Portals to Advance Scientific Research and Drug Discovery

CAMBRIDGE, Mass., April 14, 2026 — Insilico Medicine, a pioneering clinical-stage biotechnology company leveraging generative artificial intelligence (AI) to revolutionize scientific research and drug discovery, has announced a significant expansion of its MMAI Gym platform. This foundational-model training framework is now enhanced with three comprehensive benchmark leaderboard portals designed to rigorously evaluate and improve AI systems used in scientific and pharmaceutical domains. This expansion marks an important milestone in standardizing AI evaluation metrics, fostering transparent comparison, and accelerating innovation within the life sciences.

The MMAI Gym serves a dual purpose: it acts both as a training environment and an evaluation benchmark, allowing organizations to develop domain-specific AI models tailored for sophisticated scientific reasoning. These models undergo systematic performance assessments across a spectrum of meticulously curated real-world tasks, ensuring relevance and practical application. By integrating training with benchmarking under one cohesive platform, Insilico Medicine creates a robust infrastructure that ensures AI systems evolve with precision to meet complex scientific challenges.

Within the broadened MMAI Gym ecosystem, three distinct benchmark leaderboard portals have been launched. The first, ScienceAI Bench, encompasses broad scientific reasoning tasks that span multiple disciplines including biology, chemistry, longevity research, materials science, and agriculture. This multifaceted approach is designed to test AI models for versatility and adaptability across a wide range of scientific inquiries, offering a holistic evaluation of machine reasoning capabilities.

The second portal, Drug Discovery Benchmark (DDB), zeroes in on end-to-end drug discovery workflows. It covers critical components such as target identification, molecular design, and optimization processes. By simulating and scoring AI performance on these tasks, DDB offers pharmaceutical researchers a transparent and objective method to compare model efficacy and reliability, effectively bridging the gap between AI innovation and practical drug development needs.

The third portal, Insilico Bench, features proprietary benchmarking tasks devised by Insilico Medicine itself, focused on tackling complex scientific challenges specifically within the realm of drug discovery. One prominent example is TargetBench, a benchmark framework created to assess the precision of AI models in target identification. Notably, Insilico’s flagship target discovery model, TargetPro, was validated using TargetBench, with supporting research published in the journal Scientific Reports, underscoring the credibility and scientific rigor embedded within the platform.

MMAI Gym’s strength lies in its synthesis of curated industry-standard benchmarks with proprietary datasets, many of which originate from experimental laboratory data. This combination ensures that AI model evaluations are grounded in real-world scientific conditions, resulting in performance metrics that reflect true utility and predictive accuracy. Moreover, the intersecting benchmark categories allow for multidimensional analysis, enabling researchers to assess models not only for breadth but also for specialized depth in their performance.

To promote transparency and broader community engagement, Insilico Medicine has made all three leaderboard portals publicly accessible. These portals initially cover over 200 benchmark tasks with plans for continual expansion. Public leaderboards pave the way for a more open scientific discourse, fostering an ecosystem where AI models can be continuously tested, refined, and compared in a fair and standardized environment, accelerating the safe adoption of AI technologies across pharmaceutical research and development.

Alex Zhavoronkov, CEO of Insilico Medicine, explained that MMAI Gym represents a transformative approach to scientific AI. “We have created a unified system that not only trains but also evaluates and continuously enhances AI models for scientific applications,” he said. “By establishing standardized benchmarks and comprehensive training regimes, we empower the scientific community with scalable, trustworthy AI tools that can genuinely impact drug discovery pipelines.”

This initiative reflects a paradigm shift toward creating standardized, scalable frameworks to train, benchmark, and compare scientific AI models universally. Historically, disparity in evaluation criteria and fragmented datasets slowed AI adoption in computational biology and chemistry, limiting translational success. The MMAI Gym framework addresses this fragmentation by providing a centralized platform designed for iterative improvement and interoperability across different AI methodologies.

Prior research by Insilico Medicine demonstrated that foundation models trained within the MMAI Gym framework showcased as much as a tenfold increase in performance on pivotal drug discovery benchmarks compared to generic foundation models. Moreover, it highlighted the inadequacies of general-purpose AI systems, which failed to deliver reliable predictions on 75–95% of evaluated tasks within these specialized scientific challenges. These results emphasize the critical importance of domain-specific model training and benchmarking.

In a landmark collaboration, Insilico and Liquid AI jointly introduced LFM2-2.6B-MMAI (v0.2.1), the first AI model trained under the MMAI Gym’s rigorous protocols. This model, despite its relatively lightweight architecture designed for efficient on-premise deployment, achieved state-of-the-art (SOTA) performance across several essential drug discovery benchmarks. The methodology and results were thoroughly documented and accepted for presentation at the International Conference on Learning Representations (ICLR) 2026, underscoring the technical prowess and industry recognition of this work.

Insilico Medicine continues to drive innovation by fusing advanced AI with automated laboratory processes, accelerating drug development timelines, and targeting previously intractable diseases, including fibrosis, oncology, immunology, metabolic disorders, and chronic pain. Their comprehensive application of Pharma.AI extends beyond pharmaceuticals into advanced materials, agriculture, nutritional supplements, and veterinary medicine, illustrating the broad utility of AI-driven scientific innovation.

Founded with the mission to extend healthy longevity through technological breakthroughs in AI-assisted biology, Insilico Medicine went public on the Hong Kong Stock Exchange in late 2025 (stock code: 03696.HK), marking a new era in the intersection of biotech and artificial intelligence. Through platforms like MMAI Gym, the company is setting new standards for how artificial intelligence can be systematically trained, tested, and deployed within scientific research, creating a blueprint for the future of AI-augmented drug discovery.

As AI technologies advance and become increasingly complex, initiatives like MMAI Gym are instrumental in ensuring that scientific models are not only accurate but also transparent and reproducible. By providing a shared platform that benchmarks both generalist and specialist AI systems, Insilico Medicine is fostering collaboration and competition that drives continual improvement, ultimately translating computational power into tangible therapeutic breakthroughs.

News Publication Date:
April 14, 2026

Web References:
https://scienceaibench.insilico.com/
https://ddb.insilico.com/
https://insilicobench.insilico.com/
https://www.nature.com/articles/s41598-026-47765-3

References:
Zhavoronkov, A., et al. (2026). Validation of TargetPro using TargetBench. Scientific Reports.
Insilico Medicine & Liquid AI (2026). LFM2-2.6B-MMAI training and benchmarking accepted at ICLR 2026.