diff --git a/.github/workflows/benchmark.yml b/.github/workflows/benchmark.yml
new file mode 100644
index 000000000..016733057
--- /dev/null
+++ b/.github/workflows/benchmark.yml
@@ -0,0 +1,29 @@
+name: Benchmark this PR
+
+on:
+  pull_request:
+    branches: [master]
+    paths-ignore: ['docs/**']
+
+permissions:
+  pull-requests: write
+  contents: read
+
+jobs:
+  benchmark:
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    strategy:
+      fail-fast: false
+      matrix:
+        version: ["1", "lts"]
+    env:
+      # Force consistent Julia depot path for self-hosted runners
+      JULIA_DEPOT_PATH: ~/.julia
+    steps:
+      - uses: MilesCranmer/AirspeedVelocity.jl@action-v1
+        with:
+          julia-version: ${{ matrix.version }}
+          script: "benchmark/benchmarks.jl"
+          extra-pkgs: "StableRNGs,Symbolics"
+          job-summary: true
diff --git a/benchmark/benchmarks.jl b/benchmark/benchmarks.jl
new file mode 100644
index 000000000..6e6319bd6
--- /dev/null
+++ b/benchmark/benchmarks.jl
@@ -0,0 +1,79 @@
+using DataDrivenDiffEq, BenchmarkTools
+using StableRNGs, LinearAlgebra
+using Symbolics
+
+const SUITE = BenchmarkGroup()
+const rng = StableRNG(123)
+
+# Synthetic data from a linear system
+n_states = 5
+n_t = 500
+X = rand(rng, n_states, n_t)
+DX = rand(rng, n_states, n_t)
+t = collect(range(0.0, 10.0, length = n_t))
+U = rand(rng, 2, n_t)
+
+# =============================================================================
+# Problem construction
+# =============================================================================
+
+SUITE["problem"] = BenchmarkGroup()
+
+SUITE["problem"]["continuous_t"] = @benchmarkable ContinuousDataDrivenProblem(
+    $X, $t
+)
+SUITE["problem"]["continuous_dx"] = @benchmarkable ContinuousDataDrivenProblem(
+    $X, $DX
+)
+SUITE["problem"]["continuous_with_control"] = @benchmarkable ContinuousDataDrivenProblem(
+    $X, $t, $U
+)
+SUITE["problem"]["discrete"] = @benchmarkable DiscreteDataDrivenProblem($X, $t)
+SUITE["problem"]["direct"] = @benchmarkable DirectDataDrivenProblem($X, $DX)
+SUITE["problem"]["kernel_interp"] = @benchmarkable ContinuousDataDrivenProblem(
+    $X, $t, GaussianKernel()
+)
+
+# =============================================================================
+# Basis construction
+# =============================================================================
+
+SUITE["basis"] = BenchmarkGroup()
+
+@variables bvars[1:5]
+SUITE["basis"]["polynomial_basis"] = @benchmarkable polynomial_basis(
+    $(collect(bvars)), 3
+)
+SUITE["basis"]["monomial_basis"] = @benchmarkable monomial_basis(
+    $(collect(bvars)), 3
+)
+SUITE["basis"]["fourier_basis"] = @benchmarkable fourier_basis(
+    $(collect(bvars)), 2
+)
+SUITE["basis"]["Basis_construct"] = @benchmarkable Basis(
+    $(sin.(bvars .+ 1)), $(collect(bvars))
+)
+
+# =============================================================================
+# Data collation and shrinkage
+# =============================================================================
+
+SUITE["collocate"] = BenchmarkGroup()
+
+SUITE["collocate"]["collocate_data"] = @benchmarkable collocate_data(
+    $X, $t, EpanechnikovKernel()
+)
+SUITE["collocate"]["optimal_shrinkage"] = @benchmarkable optimal_shrinkage($X)
+SUITE["collocate"]["optimal_shrinkage!"] = @benchmarkable optimal_shrinkage!(
+    $(copy(X))
+)
+
+# =============================================================================
+# Normalization (DataNormalization is a processing spec used in problem options)
+# =============================================================================
+
+SUITE["normalize"] = BenchmarkGroup()
+
+SUITE["normalize"]["problem_with_normalize"] = @benchmarkable ContinuousDataDrivenProblem(
+    $X, $t, GaussianKernel(); normalize = DataNormalization()
+)