← All use cases Exploring

Solver engines

Faster solves, from heuristics proven on your test set.

For companies with a MIP, SAT or constraint solver inside their product.

A solver’s speed comes from many small rules: which variable to branch on, which cuts to keep, when to dive for a good answer. Run times are noisy, so a rule that looks faster on one run often is not. We have not run Argmax on a solver engine yet. If this is your code, talk to us about being first.

What we work on
  • Branching and variable-selection rules
  • Cut selection and primal (diving) heuristics
  • The core search loop of a SAT solver
What “better” means
  • Solve time (for example PAR-2)
  • Problems solved within the time limit
  • Primal gap at a fixed time
Proof so far
  • NVIDIA (research, 2025) AI-evolved SAT solvers solved 347 test problems, against 334 for the winner of the 2025 SAT Competition. Source
  • DHEvo (research, 2025) AI-written diving heuristics for the SCIP solver left about half the primal gap of an earlier AI method on set cover (9.74% against 20.39%). Source
  • MILP-Evo (research, 2026) Evolved cut and branching rules made SCIP 74–84% faster on one problem family, and slower on the hardest instances of others. Gains on some problems do not carry over to all. Source

Research results, not Argmax results. They show that solver heuristics have room to improve, and that a gain must be checked on problems the AI never saw.

The checks are the same everywhere

Old and new code run on the same problems, several times each. A statistical test rules out luck and nothing else may get worse. Some problems stay out of the AI’s reach; results on those are measured and reported too. Every accepted change arrives as a pull request your engineers review.

Book a call → How it works

Other use cases

Open for a pilot

Supply-chain planning

For companies that build planning software: forecasting, inventory, production planning.

See the use case → Open for a pilot

Machine & job-shop scheduling

For companies that build scheduling software for factories: jobs on machines.

See the use case → Open for a pilot

Packing & cutting

For companies whose software packs or cuts: bins, containers, pallets, sheets, rolls.

See the use case → Exploring

Energy planning

For companies that build planning software for power plants, batteries and EV charging.

See the use case → Exploring

Beyond optimisation

For teams whose code has a speed or size objective: compilers, databases, ML training.

See the use case →