grbcompat

A Python package that lets you run gurobipy-based optimization models on the free, open-source HiGHS solver — with no changes to your existing code.

Why

Gurobi is a best-in-class commercial solver, but it requires a licence. HiGHS is a high-quality open-source LP/MIP solver. This wrapper translates the gurobipy API into HiGHS calls at runtime, so you can:

• Run models on HiGHS during development without a Gurobi licence.
• Switch freely between solvers to compare results or costs.
• Keep a single codebase that works with either solver.

Requirements

• Python ≥ 3.11
• highspy ≥ 1.7 (installed automatically)
• A Gurobi licence is not required to use this wrapper.

Installation

pip install grbcompat

Or, to work from source:

git clone <repo-url>
cd grbcompat
uv sync --dev

---

Usage

Pattern A — Drop-in replacement (one extra line)

Add a single install() call before any import gurobipy statement. Every subsequent import gurobipy in that process resolves to the HiGHS-backed wrapper. Your existing code stays untouched.

import grbcompat
grbcompat.install()          # ← the only change

import gurobipy as gp                # now backed by HiGHS
from gurobipy import GRB

m = gp.Model("my_model")
x = m.addVar(lb=0, name="x")
y = m.addVar(lb=0, name="y")

m.setObjective(x + 2 * y, GRB.MINIMIZE)
m.addConstr(x + y >= 1, "demand")
m.optimize()

print(f"x={x.X:.4f}  y={y.X:.4f}  obj={m.ObjVal:.4f}")

Pattern B — Side-by-side solvers

Import the wrapper directly under its own name. Both solver namespaces are fully independent and can be used in the same script simultaneously.

import grbcompat as highs    # backed by HiGHS (free)
import gurobipy as gp                # backed by Gurobi (licensed)

from grbcompat import GRB as HGRB

# Build the same LP with both solvers
def build(Mod, G):
    m = Mod()
    x = m.addVar(lb=0.0, name="x")
    y = m.addVar(lb=0.0, name="y")
    m.addConstr(x + y >= 4, "demand")
    m.setObjective(2 * x + 3 * y, G.MINIMIZE)
    m.optimize()
    return m, x, y

m_h, x_h, y_h = build(highs.Model, HGRB)
m_g, x_g, y_g = build(gp.Model, gp.GRB)

print(f"HiGHS:  obj={m_h.ObjVal:.4f}  x={x_h.X:.4f}  y={y_h.X:.4f}")
print(f"Gurobi: obj={m_g.ObjVal:.4f}  x={x_g.X:.4f}  y={y_g.X:.4f}")

---

Supported gurobipy API

Model

Method / Property	Notes
addVar(lb, ub, obj, vtype, name)	All types: CONTINUOUS, INTEGER, BINARY
addVars(*indices, lb, ub, obj, vtype, name)	Scalar, list, range, or cartesian-product indices; lb/ub can be dicts
addConstr(lhs, sense, rhs, name)	TempConstr from <=/>=/==, or explicit sense/rhs
addConstrs(generator, name)	Returns a tupledict
addLConstr(...)	Alias for addConstr
setObjective(expr, sense)	GRB.MINIMIZE / GRB.MAXIMIZE
optimize()
update()	No-op (changes applied immediately)
reset()	Clears solution without rebuilding the model
write(filename)	Writes .lp, .mps, or other HiGHS-supported formats
setParam(name, value)	Alias for model.Params.<name> = value
setAttr(attr, value)	Supports ModelSense, ObjCon
getAttr(attr, objects)	Batch-reads X, Pi, Slack, RC, LB, UB, …
getVars() / getConstrs()
getVarByName(name) / getConstrByName(name)
ObjVal, ObjBound, MIPGap	After optimize()
Status, Runtime
NumVars, NumConstrs
ModelName, ModelSense	Readable and writable
Context manager (with Model() as m)

Var

Attribute / Property	Notes
X	Primal value (after optimize())
RC	Reduced cost
LB, UB, Obj, VType, VarName	Readable and writable; changes take effect immediately
+, -, *, /, unary -	Returns LinExpr
<=, >=, ==	Returns TempConstr for use in addConstr

Constr

Attribute / Property	Notes
Pi	Dual value / shadow price (after optimize())
Slack	Constraint slack (after optimize())
ConstrName, Sense, RHS	RHS is writable; change takes effect at next optimize()

LinExpr

Full arithmetic support: +, -, *, /, unary -, sum() built-in, ==/<=/>= to produce constraints.
Methods: size(), getCoeff(i), getVar(i), getConstant(), add(), addTerms(), getValue().

tupledict

Returned by addVars and addConstrs. Supports select(*pattern), sum(*pattern), prod(coeff_dict, *pattern).

GRB constants

GRB.MINIMIZE / GRB.MAXIMIZE
GRB.CONTINUOUS / GRB.INTEGER / GRB.BINARY / GRB.SEMICONT / GRB.SEMIINT
GRB.LESS_EQUAL / GRB.GREATER_EQUAL / GRB.EQUAL
GRB.INFINITY
GRB.OPTIMAL / GRB.INFEASIBLE / GRB.UNBOUNDED / GRB.INF_OR_UNBD
GRB.TIME_LIMIT / GRB.ITERATION_LIMIT / GRB.SOLUTION_LIMIT / …
GRB.Status.*   # mirrors the top-level codes
GRB.Attr.*     # attribute name strings

Params

Set solver parameters via attribute assignment. gurobipy names are translated to their HiGHS equivalents automatically.

gurobipy name	HiGHS option	Type
TimeLimit	time_limit	float
MIPGap	mip_rel_gap	float
MIPGapAbs	mip_abs_gap	float
FeasibilityTol	primal_feasibility_tolerance	float
OptimalityTol	dual_feasibility_tolerance	float
IntFeasTol	mip_feasibility_tolerance	float
OutputFlag	output_flag	bool
Threads	threads	int
Seed	random_seed	int
NodeLimit	mip_max_nodes	int
IterationLimit	simplex_iteration_limit	int

Unknown parameter names are forwarded directly to HiGHS.

Compatibility stubs

GurobiError, Env, disposeDefaultEnv — accepted and silently ignored so that licence-management code in existing scripts does not break.

Module-level functions

grbcompat.quicksum(iterable)      # → LinExpr
grbcompat.multidict(data)         # → [keys, dict1, dict2, …]
grbcompat.install()               # patches sys.modules['gurobipy']

---

Running the tests

uv sync --dev
uv run pytest

The test suite has 388 tests across 10 files:

File	Tests	Coverage
test_constants.py	28	All GRB constants and status codes
test_expr.py	47	LinExpr arithmetic, merging, TempConstr
test_var.py	52	All Var properties, setters, operators
test_constr.py	22	Constr metadata, duals, slacks, RHS setter
test_tupledict.py	31	select, sum, prod, addVars integration
test_params.py	16	All mapped params, unknown-param error
test_model_core.py	81	addVar/Vars, addConstr/Constrs, setObjective, I/O
test_model_lp.py	32	LP correctness, duals, re-optimization
test_model_mip.py	15	Binary, integer, mixed-integer programs
test_api_compat.py	55	install() isolation, gurobipy usage patterns
test_gurobi_comparison.py	29	Cross-solver comparison (skipped without Gurobi licence)

Cross-solver comparison tests

If you have a valid Gurobi licence, the comparison tests run automatically and verify that HiGHS and Gurobi produce identical results (within 1e-4 tolerance) for LP objectives, primal values, dual values, slacks, solve status, and MIP objectives:

pytest tests/test_gurobi_comparison.py -v

Without a licence they are skipped with a clear message — no configuration needed.

---

Limitations

• Quadratic objectives and constraints (QuadExpr, QConstr) are not supported. HiGHS does support QP but the translation layer is not yet implemented.
• Model.remove() (removing individual variables or constraints) is not implemented.
• Callbacks (Model.optimize(callback)) are not supported.
• Multi-objective and scenario features are not supported.
• HiGHS and Gurobi may return different optimal bases for degenerate problems; primal solution values can differ while objective values agree.

---

How it works

When you call install(), the package registers itself under sys.modules['gurobipy']. Subsequent import gurobipy statements in the same process return the wrapper instead of the real package.

Internally, each Model creates a highspy.Highs instance. Variable and constraint operations translate directly to HiGHS column/row operations applied immediately (no lazy batching). model.update() is a no-op.

The solver-status integer codes (GRB.OPTIMAL = 2, etc.) match gurobipy's values exactly, so any code that branches on model.Status works correctly with either solver.

---

License

MIT