diff --git a/asv_bench/benchmarks/groupby.py b/asv_bench/benchmarks/groupby.py
index 0f37933a1b768..1648985a56b91 100644
--- a/asv_bench/benchmarks/groupby.py
+++ b/asv_bench/benchmarks/groupby.py
@@ -507,11 +507,11 @@ def time_frame_agg(self, dtype, method):
         self.df.groupby("key").agg(method)
 
 
-class CumminMax:
+class Cumulative:
     param_names = ["dtype", "method"]
     params = [
         ["float64", "int64", "Float64", "Int64"],
-        ["cummin", "cummax"],
+        ["cummin", "cummax", "cumsum"],
     ]
 
     def setup(self, dtype, method):
diff --git a/pandas/_libs/groupby.pyx b/pandas/_libs/groupby.pyx
index b72b927b3c2a8..c05dbf5e3c8ec 100644
--- a/pandas/_libs/groupby.pyx
+++ b/pandas/_libs/groupby.pyx
@@ -247,24 +247,24 @@ def group_cumsum(numeric[:, ::1] out,
             for j in range(K):
                 val = values[i, j]
 
+                # For floats, use Kahan summation to reduce floating-point
+                # error (https://en.wikipedia.org/wiki/Kahan_summation_algorithm)
                 if numeric == float32_t or numeric == float64_t:
                     if val == val:
                         y = val - compensation[lab, j]
                         t = accum[lab, j] + y
                         compensation[lab, j] = t - accum[lab, j] - y
                         accum[lab, j] = t
-                        out[i, j] = accum[lab, j]
+                        out[i, j] = t
                     else:
                         out[i, j] = NaN
                         if not skipna:
                             accum[lab, j] = NaN
                             break
                 else:
-                    y = val - compensation[lab, j]
-                    t = accum[lab, j] + y
-                    compensation[lab, j] = t - accum[lab, j] - y
+                    t = val + accum[lab, j]
                     accum[lab, j] = t
-                    out[i, j] = accum[lab, j]
+                    out[i, j] = t
 
 
 @cython.boundscheck(False)