Mathematical optimization depends on solvers - yet using them effectively can be daunting. At GAMS, we turn solver complexity into solver power, providing both a unified modeling interface and deep, decades-long expertise in solver development.
From Solver Complexity to Solver Independence
Optimization solvers are powerful but often intimidating tools. When used directly through their APIs, they expose a wide range of configuration options, parameters, and performance settings that require deep technical understanding. Each solver has its own conventions, syntax, and option names, making it difficult to transfer knowledge from one solver to another. Understanding how these settings influence results takes significant expertise, and switching between solvers becomes a time-consuming and error-prone task. By contrast, using solvers through a modeling language like GAMS or GAMSPy abstracts away these differences, allowing users to focus on formulating their optimization problems rather than dealing with solver-specific details.
This abstraction has practical consequences: switching from one solver to another takes a single line of code, making experimentation with different algorithms such as simplex and barrier as well as various fine-tuning option settings straightforward in GAMS and GAMSPy.
What it Means to Truly Understand Solvers
At GAMS, solvers are not a black box. We understand them in depth because we have worked closely with all major commercial and open-source solver developers for decades. This collaboration goes beyond simple integration. We build and maintain the interfaces that connect GAMS to each solver, giving us hands-on insight into how they work and how to make them perform at their best.
Our quality assurance process reinforces this expertise. Every night, we run thousands of tests with all supported combinations of solvers as part of our automated testing pipeline. This ensures that our interfaces remain robust, efficient, and consistent, especially as solvers evolve and new ones emerge.
Behind this effort is a team of PhD-level researchers and engineers who specialize in numerical optimization. They are responsible for maintaining solver integrations and providing direct technical support to customers who encounter solver-specific challenges. When complex issues arise, our team can rely on established communication channels with solver developers to resolve them quickly, often accessing insights that are not publicly documented.
Some of our staff members have their roots at leading centers for optimization research, such as the Zuse Institute Berlin and the University of Wisconsin-Madison. Combined with extensive experience across industries such as engineering, economics, energy, and finance, this background positions GAMS uniquely. Our customers benefit from a rare combination of deep theoretical knowledge and decades of applied expertise in using solvers effectively.
Beyond Integration: GAMS’ Role in Solver Innovation
While GAMS integrates and supports nearly all major solvers equally, we also contribute to the solver ecosystem itself. Our involvement in both open source and commercial solver projects like CONOPT and PATH reflects our technical commitment to advancing solver technology, not a commercial preference.
Commercial Solvers We Develop and Support
CONOPT is a leading nonlinear programming (NLP) solver that became part of GAMS in 2024. By bringing CONOPT fully in-house, we ensured a smooth transition from its original developer, Arne Drud from ARKI Consulting, and secured its ongoing development at GAMS. CONOPT has long been a trusted solver in computable general equilibrium models, engineering applications, and process optimization. More recently, it has shown strong potential in emerging areas that combine machine learning with optimization. This is an area the GAMSPy team at GAMS is actively driving forward.
PATH is a solver designed for mixed complementarity problems. Steve Dirkse, President of GAMS Development Corp, is one of PATH’s original developers. The solver has earned a reputation as one of the most reliable and efficient tools in the field of Computable General Equilibrium (CGE) modeling and related applications and Steve was awarded the Beale-Orchard-Hayes Prize in 1997 for his work on PATH, together with Michael C Ferris.
Open Source Solvers We Help Advance
GAMS has been a long time supporter of the COIN-OR foundation, an open-source community for operations research software. Beginning in 2004 with a shared fascination for simple branch-and-bound solvers, GAMS has from early times on contributed to broaden the user base of COIN-OR solvers BONMIN, CBC, CLP, COUENNE, and IPOPT by making them easily available via the GAMS distribution, an effort that was acknowledged by awarding the COIN-OR Cup 2012 jointly to AIMMS, GAMS, and MPL. CLP (COIN-OR Linear Programming) and CBC (COIN-OR Branch-and-Cut) are linear and mixed-integer linear programming solvers, developed by operations research veteran John Forrest. IPOPT (Interior Point OPTimizer) on the other hand is a large-scale nonlinear programming solver that implements an interior-point line-search filter method, developed primarily by Andreas Wächter. On top of CBC and IPOPT build the MINLP solvers BONMIN and COUENNE , developed by Pierre Bonami, Pietro Belotti, and others. GAMS has made its links to these and other open-source solvers available within COIN-OR, serving now as guidelines for solver developers to connect their solvers to GAMS. GAMS has also contributed to newer arrivals in the COIN-OR ecosystem, in particular HiGHS (for LP/MIP) and SHOT (MINLP). For many years now, the support of GAMS in maintaining some of the COIN-OR solvers helped ensure users can keep leveraging their powerful capabilities.
As an alternative to the COIN-OR ecosystem, SCIP (Solving Constraint Integer Programs) is a versatile and powerful open-source solver for mixed-integer programming (MIP), mixed-integer nonlinear programming (MINLP), and constraint programming (CP) problems. The steady development of the SCIP Optimization Suite in a cooperation of, currently, eight academic institutions and several solvers vendors, including Cardinal, FICO, GAMS, and Gurobi, has led to an extremely flexible and feature-rich framework for various optimization algorithms, which includes some of the fastest non-commercial solvers available today. GAMS has contributed to the development of SCIP for more than 10 years and has recently increased this investment even further, being proud to now account for 3 SCIP developers in the GAMS development team. With PaPILO, SCIP, and SoPlex included in the GAMS distribution, these solvers are readily available for our users to serve as valuable tools to solve a broad spectrum of discrete and continuous optimization problems.
The new kid on the block - GPU-Powered PDHG
An exciting and very recent addition to the solver landscape is the primal-dual hybrid gradient algorithm for LPs, especially an improved version called PDLP, which stems from research by Google’s OR-Tools team . This method is well-suited (by design) to run in parallel on modern GPUs, in contrast to established LP methods like simplex or interior point methods. PDHG has been integrated quickly into solvers like HiGHS, COPT, Gurobi, and KNITRO. Nvidia has also entered the fray and released cuOpt (also in COIN-OR), a GPU-accelerated implementation inspired by PDHG, which is tuned for the latest Nvidia GPU hardware. During a fruitful collaboration with Nvidia we were able to quickly integrate cuOpt into GAMS and make this available to anyone interested in this promising new technology. The hardware needed is still extremely expensive and will stop most potential users from exploring the benefits of the technology. However, once the required GPUs are offered by the big hyper-scalers, we will integrate this into our Engine-SaaS deployment solution and make this technology available to a much wider audience.
Those examples highlight the unique position of GAMS in the optimization landscape - not just as a platform that integrates solvers, but as an active contributor to solver development and innovation, resulting in an inside perspective on how modern solvers work.
How Our Solver Expertise Translates into Customer Success
Our deep understanding of solvers directly translates into value for our customers. When users face difficulties getting the most out of a solver, our experts can quickly pinpoint issues, explain solver behavior, and suggest effective parameter settings or model adjustments. This support often determines whether a model converges slowly or delivers reliable, fast results.
For example, in collaboration with Austrian Power Grid (APG), our consulting team improved solver stability and efficiency in a large scale energy optimization model, reducing total solve time by up to 70% and memory use by nearly 80%. In another case, working with TotalEnergies, we restructured a complex MINLP model for carbon storage, cutting solve times from hours to minutes and creating a user-friendly Python + Excel interface that brought advanced optimization directly to field engineers.
We also help customers make informed decisions when choosing between solvers. Because GAMS provides access to virtually all leading commercial and open-source solvers, we can evaluate options objectively. We have no commercial preference for one solver over another, which allows us to give independent recommendations based solely on technical merit, problem characteristics, and - where relevant - the associated licensing costs.
This unique combination of deep solver knowledge, broad solver coverage, and true independence makes GAMS a trusted partner for anyone aiming to get the most out of optimization technology.
Modeling Practices That Unlock Solver Performance
Even the best solver can only perform as well as the model allows. Many optimization issues stem not from the solver itself, but from the way a problem is formulated. Poor scaling, unnecessary nonlinearities, or the absence of decomposition strategies can pose additional challenges for any solver and lead to slow or unreliable results.
At GAMS, we see model formulation and solver performance as two sides of the same coin. Our support and consulting teams work closely with customers to identify and resolve such modeling bottlenecks. Whether through decomposition techniques, reformulations that improve numerical conditioning, or guidance on variable scaling, we help users create models that solvers can handle efficiently and robustly.
This combination of solver knowledge and modeling expertise allows us to deliver practical, performance-oriented solutions — ensuring that our customers choose the right solver and use it under the best possible conditions.
From Insight to Impact
The optimization landscape is diverse and technically demanding, and success often depends on choosing and using the right solver effectively. This is where GAMS stands out. With decades of experience, deep technical expertise, and long-standing relationships with solver vendors, we combine the advantages of independence and insight.
Our customers benefit from this unique position. They gain access not only to a wide range of solvers but also to the collective knowledge of a team that understands how these tools work at a fundamental level. Whether it is selecting the best solver for a specific model, fine-tuning performance, or troubleshooting complex behavior, GAMS provides guidance grounded in both theory and practice.
In short, GAMS transforms solver complexity into solver power—helping customers achieve better model performance, faster results, and deeper confidence in their optimization solutions. As solver technology evolves - from hybrid algorithms to GPU-accelerated methods - GAMS continues to integrate these advances into a consistent, solver-independent environment.
Interested in a discussion? Contact us at support@gams.com !