Solver Benchmarks -

⏱️ Help us benchmark CVXPY solvers! Contribute problems and results to the Solver Benchmarks project. Read our new paper on the CVXPY internals!
The solver-benchmarks project collects benchmark problems and results to inform CVXPY’s default solver selection. By running a shared suite of problems across many machines and solver versions, we can make data-driven decisions about which solvers to recommend as defaults for each problem type (LP, QP, MIP, SOCP, SDP).
CVXPY supports many solvers, but choosing good defaults requires real performance data across a wide range of problems and environments. The solver-benchmarks project provides that data so CVXPY can recommend the best solver for each problem type out of the box.
We are collecting benchmark problems from the community. Ideal problems:
Take about one second to solve.
Are based on real-world applications.
Use real data when possible.
Cover the full range of problem types: LP, QP, MIP, SOCP, SDP.
If you have a problem that fits these criteria, please contribute it!
Once the problem suite is established, contributors run benchmarks on their machines and submit results. Results are stored as JSONL files for easy comparison across machines and solver versions.
See the GitHub repository for full instructions and contribution guidelines.