It’s been one year, and what started as an idea to bring GAMS technology natively into Python has become a trusted tool for the optimization community. GAMSPy, our Python-native interface to the GAMS modeling system, has grown from a concept to a globally used tool.
With thousands of academic users, adoption at leading universities, and growing use in professional workflows, this first anniversary highlights our technical progress and community engagement. Looking ahead, the next year will focus on expanding functionality, strengthening the ecosystem, and supporting the growing number of researchers and practitioners who rely on GAMSPy.
The idea is simple, you write models in idiomatic Python, and in the background the GAMS execution system handles deterministic model generation and solves them with our portfolio of free and commercial solvers. We built it to meet modelers where they work - Python - so teaching, prototyping, and production use the same environment with no rewrites.
For academia this was a step change: for the first time, academic users can access the full power of GAMS model generation and commercial-grade solvers with free academic licenses - enabling larger student assignments and research projects.
One year in, GAMSPy has been adopted by thousands of academic users at leading universities. Initially, most adoption came from individual users obtaining licenses, but we now see clear evidence of increased classroom use. In its inaugural year, we have distributed roughly 7,500 academic GAMSPy licenses in 95 countries, with adoption at 79 of the world’s top 100 universities –clear evidence of its value in teaching and research.
For commercial teams, GAMSPy shortens time-to-value by keeping models in Python while preserving solver independence and performance, making it easier to plug optimization into existing data and CI/CD pipelines. On top, the integration with GAMS MIRO and GAMS Engine allows creating user interfaces for analysts and highly scalable deployments.
Of course, we do not yet have commercial usage numbers comparable to academia, but the appeal of GAMSPy is clear: it accelerates the path from prototype to production by keeping optimization models within Python while maintaining solver independence. Python acts as a high-level abstraction layer, delegating heavy computational tasks to the GAMS backend. Benchmarks show minimal overhead - about 27% on Linux and 8% on Windows - figures negligible in real-world cases where solver runtime dominates execution.
Machine Learning
In its first year, GAMSPy has gradually introduced more and more constructs familiar from machine learning, such as linear layers, pooling operators (max, min, average), and activation functions (e.g., Leaky ReLU). While GAMSPy is not a deep learning framework, these features provide a compact way to express neural-network-inspired structures within optimization models. This opens new avenues at the intersection of optimization and machine learning, supporting research in adversarial training, hyperparameter tuning, robust learning, and hybrid models that integrate neural components with mathematical programming.
For academic users, having ML-style operators in GAMSPy means they can formulate hybrid models that combine discrete/continuous optimization with neural-network layers, while still solving them with the full power of commercial optimization solvers - and with no license cost in teaching and research. For commercial teams, these features open the door to decision-focused learning or model compression problems where optimization and ML intersect. In short, the ML functions extend GAMSPy beyond “classic” mathematical programming into the fast-growing space of optimization-aware ML and ML-augmented optimization.
Community and resources
Alongside the software, we’ve built a place where users can connect: the GAMS Forum . When questions or problems arise, GAMSPy users don’t have to work in isolation –they can post directly to the forum and get input from both the GAMS team and other experienced community members. Because the discussions are public and searchable, the forum quickly becomes a knowledge base of practical solutions, tips, and workarounds.
For academic users, this means faster answers when teaching or research deadlines are tight. For commercial users, it provides qualified second opinions and guidance on advanced modeling or integration issues. The GAMS Forum ensures that GAMSPy is more than just a resource —it is a tool backed by an active, knowledgeable community.
Beyond our forum, GAMSPy is supported by excellent documentation at https://gamspy.readthedocs.io , which offers detailed guides and API references. Plus, we also provide a great selection of practical examples on GitHub and will be developing more online teaching resources with trusted partners to integrate GAMSPy into academic and professional training.
Final thoughts
The rapid growth and adoption of GAMSPy in its first year underscore its significant impact on the mathematical optimization community. We’ve seen students and researchers develop novel and creative optimization pipelines fully integrated into Python, such as those highlighted in our first GAMSPy student competition . As we look to the future, we remain committed to enhancing GAMPy capabilities and fostering a vibrant ecosystem, empowering an ever-wider range of users to leverage the power of GAMS within Python for their research and professional endeavors. At the same time, we continue to extend and improve GAMS itself - GAMS is the workhorse underpinning GAMSPy’s performance, and will continue to have its place for our users.