Insilico Medicine has expanded its Science MMAI Gym, a large-scale training and benchmarking platform for artificial intelligence, with the launch of three public leaderboard portals designed to evaluate AI performance across scientific research and drug discovery. Positioned as both a training environment and benchmarking system, MMAI Gym enables the development of domain-specific AI models while rigorously assessing their capabilities on real-world tasks.

The newly launched benchmark categories include ScienceAI Bench, which evaluates general scientific reasoning across disciplines such as biology, chemistry, and materials science; the Drug Discovery Benchmark (DDB), focused on end-to-end pharmaceutical R&D tasks; and Insilico Bench, a proprietary suite targeting complex and emerging scientific challenges. Together, these benchmarks draw from both curated industry datasets and proprietary, experimentally grounded data, enabling multi-dimensional evaluation across more than 200 tasks.

The platform reflects a broader shift toward standardized, scalable evaluation of scientific AI systems. Previous results from Insilico demonstrate that models trained within MMAI Gym can achieve up to tenfold performance improvements on key drug discovery benchmarks. In collaboration with Liquid AI, Insilico also developed a compact foundation model that achieved state-of-the-art performance across multiple drug discovery tasks, with findings presented at ICLR 2026.

By integrating training, benchmarking, and public evaluation, MMAI Gym aims to accelerate the adoption of reliable, high-performance AI systems across pharmaceutical research and beyond.

Researchers have developed a feature selection-based solar irradiance forecasting method to improve the operation of stand-alone photovoltaic systems. The approach uses a bidirectional long short-term memory hybrid network to forecast solar irradiance and then applies the forecasted data to estimate the optimum tilt angle of photovoltaic panels, helping increase PV output power.

For the first time, scientists have directly imaged the quantum process underlying superconductivity, a phenomenon in which paired electrons cause electric current to flow without resistance at sufficiently low temperatures. The results weren’t quite what they expected. In the study, published April 15 in Physical Review Letters, the scientists directly imaged individual atoms pairing up in a special gas cooled nearly to absolute zero — the unreachable limit to how cold things can get. The type of gas, called a Fermi gas, allows scientists to substitute electrons with atoms and probe the physics of superconductors in a controlled way. Surprisingly, the scientists found that after pairing up, the atoms moved in a synchronized dance, with their positions dependent on those of other pairs — a phenomenon not predicted by the 70-year-old, Nobel-prize-winning theory of superconductivity.

A key question in physics is whether gravity follows quantum rules, but testing this is difficult because gravitational effects are so weak. Researchers from Kyushu University have theoretically proposed a method using momentum-squeezed states in optomechanical systems to amplify gravity-induced entanglement signals. This approach could make such signals easier to detect, paving the way for future experiments to determine whether gravity has a quantum nature.