diff --git a/base/Base.jl b/base/Base.jl
index 63728fdba3e4e..f8df4047eb6fa 100644
--- a/base/Base.jl
+++ b/base/Base.jl
@@ -103,6 +103,17 @@ include("generator.jl")
 include("reflection.jl")
 include("options.jl")
 
+# define invoke(f, T, args...; kwargs...), without kwargs wrapping
+# to forward to invoke
+function Core.kwcall(kwargs, ::typeof(invoke), f, T, args...)
+    @inline
+    # prepend kwargs and f to the invoked from the user
+    T = rewrap_unionall(Tuple{Any, Core.Typeof(f), (unwrap_unionall(T)::DataType).parameters...}, T)
+    return invoke(Core.kwcall, T, kwargs, f, args...)
+end
+# invoke does not have its own call cache, but kwcall for invoke does
+typeof(invoke).name.mt.max_args = 3 # invoke, f, T, args...
+
 # core operations & types
 include("promotion.jl")
 include("tuple.jl")
diff --git a/base/boot.jl b/base/boot.jl
index 85e50b626185f..80ef23cd0fd78 100644
--- a/base/boot.jl
+++ b/base/boot.jl
@@ -369,11 +369,11 @@ include(m::Module, fname::String) = ccall(:jl_load_, Any, (Any, Any), m, fname)
 
 eval(m::Module, @nospecialize(e)) = ccall(:jl_toplevel_eval_in, Any, (Any, Any), m, e)
 
+# dispatch token indicating a kwarg (keyword sorter) call
 function kwcall end
+# deprecated internal functions:
 kwfunc(@nospecialize(f)) = kwcall
 kwftype(@nospecialize(t)) = typeof(kwcall)
-kwcall(kwargs, ::typeof(invoke), f, T, args...) = kwinvoke(kwargs, invoke, f, T, args...)
-# invoke(f, T, args...; kwargs...) = kwinvoke(kwargs, invoke, f, T, args...)
 
 mutable struct Box
     contents::Any
diff --git a/src/builtin_proto.h b/src/builtin_proto.h
index d2fa595013d59..64e3fbd1af366 100644
--- a/src/builtin_proto.h
+++ b/src/builtin_proto.h
@@ -42,7 +42,6 @@ DECLARE_BUILTIN(is);
 DECLARE_BUILTIN(isa);
 DECLARE_BUILTIN(isdefined);
 DECLARE_BUILTIN(issubtype);
-DECLARE_BUILTIN(kwinvoke);
 DECLARE_BUILTIN(modifyfield);
 DECLARE_BUILTIN(nfields);
 DECLARE_BUILTIN(setfield);
diff --git a/src/builtins.c b/src/builtins.c
index 8e56b25a2c6ca..323a42b91ca92 100644
--- a/src/builtins.c
+++ b/src/builtins.c
@@ -1359,50 +1359,6 @@ JL_CALLABLE(jl_f_invoke)
     return res;
 }
 
-JL_CALLABLE(jl_f_kwinvoke)
-{
-    JL_NARGSV(invoke, 3);
-    jl_value_t *kwargs = args[0];
-    // args[1] is `invoke` itself
-    jl_value_t *func = args[2];
-    jl_value_t *argtypes = args[3];
-    jl_value_t *kws = jl_kwcall_func;
-    JL_GC_PUSH1(&argtypes);
-    if (jl_is_tuple_type(argtypes)) {
-        // construct a tuple type for invoking a keyword sorter by putting the kw container type
-        // and the type of the function at the front.
-        size_t i, nt = jl_nparams(argtypes) + 2;
-        if (nt < jl_page_size/sizeof(jl_value_t*)) {
-            jl_value_t **types = (jl_value_t**)alloca(nt*sizeof(jl_value_t*));
-            types[0] = (jl_value_t*)jl_namedtuple_type;
-            types[1] = jl_is_type(func) ? (jl_value_t*)jl_wrap_Type(func) : jl_typeof(func);
-            for (i = 2; i < nt; i++)
-                types[i] = jl_tparam(argtypes, i - 2);
-            argtypes = (jl_value_t*)jl_apply_tuple_type_v(types, nt);
-        }
-        else {
-            jl_svec_t *types = jl_alloc_svec_uninit(nt);
-            JL_GC_PUSH1(&types);
-            jl_svecset(types, 0, jl_namedtuple_type);
-            jl_svecset(types, 1, jl_is_type(func) ? (jl_value_t*)jl_wrap_Type(func) : jl_typeof(func));
-            for (i = 2; i < nt; i++)
-                jl_svecset(types, i, jl_tparam(argtypes, i - 2));
-            argtypes = (jl_value_t*)jl_apply_tuple_type(types);
-            JL_GC_POP();
-        }
-    }
-    else {
-        // invoke will throw an error
-    }
-    args[0] = kws;
-    args[1] = argtypes;
-    args[2] = kwargs;
-    args[3] = func;
-    jl_value_t *res = jl_f_invoke(NULL, args, nargs);
-    JL_GC_POP();
-    return res;
-}
-
 // Expr constructor for internal use ------------------------------------------
 
 jl_expr_t *jl_exprn(jl_sym_t *head, size_t n)
@@ -2011,7 +1967,6 @@ void jl_init_primitives(void) JL_GC_DISABLED
     // method table utils
     jl_builtin_applicable = add_builtin_func("applicable", jl_f_applicable);
     jl_builtin_invoke = add_builtin_func("invoke", jl_f_invoke);
-    jl_builtin_kwinvoke = add_builtin_func("kwinvoke", jl_f_kwinvoke);
 
     // internal functions
     jl_builtin_apply_type = add_builtin_func("apply_type", jl_f_apply_type);
diff --git a/src/codegen.cpp b/src/codegen.cpp
index 8477c3c9c1427..b00a07e664a68 100644
--- a/src/codegen.cpp
+++ b/src/codegen.cpp
@@ -1169,7 +1169,6 @@ static const auto &builtin_func_map() {
           { jl_f_svec_addr,               new JuliaFunction{XSTR(jl_f_svec), get_func_sig, get_func_attrs} },
           { jl_f_applicable_addr,         new JuliaFunction{XSTR(jl_f_applicable), get_func_sig, get_func_attrs} },
           { jl_f_invoke_addr,             new JuliaFunction{XSTR(jl_f_invoke), get_func_sig, get_func_attrs} },
-          { jl_f_kwinvoke_addr,           new JuliaFunction{XSTR(jl_f_kwinvoke), get_func_sig, get_func_attrs} },
           { jl_f_isdefined_addr,          new JuliaFunction{XSTR(jl_f_isdefined), get_func_sig, get_func_attrs} },
           { jl_f_getfield_addr,           new JuliaFunction{XSTR(jl_f_getfield), get_func_sig, get_func_attrs} },
           { jl_f_setfield_addr,           new JuliaFunction{XSTR(jl_f_setfield), get_func_sig, get_func_attrs} },
diff --git a/src/staticdata.c b/src/staticdata.c
index 0952331ba6354..c1ae4bb249b51 100644
--- a/src/staticdata.c
+++ b/src/staticdata.c
@@ -81,7 +81,7 @@ extern "C" {
 // TODO: put WeakRefs on the weak_refs list during deserialization
 // TODO: handle finalizers
 
-#define NUM_TAGS    159
+#define NUM_TAGS    157
 
 // An array of references that need to be restored from the sysimg
 // This is a manually constructed dual of the gvars array, which would be produced by codegen for Julia code, for C.
@@ -251,7 +251,6 @@ jl_value_t **const*const get_tags(void) {
         INSERT_TAG(jl_builtin_apply_type);
         INSERT_TAG(jl_builtin_applicable);
         INSERT_TAG(jl_builtin_invoke);
-        INSERT_TAG(jl_builtin_kwinvoke);
         INSERT_TAG(jl_builtin__expr);
         INSERT_TAG(jl_builtin_ifelse);
         INSERT_TAG(jl_builtin__typebody);
@@ -312,7 +311,7 @@ static const jl_fptr_args_t id_to_fptrs[] = {
     &jl_f_throw, &jl_f_is, &jl_f_typeof, &jl_f_issubtype, &jl_f_isa,
     &jl_f_typeassert, &jl_f__apply_iterate, &jl_f__apply_pure,
     &jl_f__call_latest, &jl_f__call_in_world, &jl_f__call_in_world_total, &jl_f_isdefined,
-    &jl_f_tuple, &jl_f_svec, &jl_f_intrinsic_call, &jl_f_kwinvoke,
+    &jl_f_tuple, &jl_f_svec, &jl_f_intrinsic_call,
     &jl_f_getfield, &jl_f_setfield, &jl_f_swapfield, &jl_f_modifyfield,
     &jl_f_replacefield, &jl_f_fieldtype, &jl_f_nfields,
     &jl_f_arrayref, &jl_f_const_arrayref, &jl_f_arrayset, &jl_f_arraysize, &jl_f_apply_type,
diff --git a/test/reduce.jl b/test/reduce.jl
index 68e0fbaf00fc0..84d93b12913e4 100644
--- a/test/reduce.jl
+++ b/test/reduce.jl
@@ -161,11 +161,13 @@ sum3(A) = reduce(plus, A)
 sum4(itr) = invoke(reduce, Tuple{Function, Any}, plus, itr)
 sum5(A) = reduce(plus, A; init=0)
 sum6(itr) = invoke(Core.kwcall, Tuple{NamedTuple{(:init,), Tuple{Int}}, typeof(reduce), Function, Any}, (init=0,), reduce, plus, itr)
+sum61(itr) = invoke(reduce, Tuple{Function, Any}, init=0, plus, itr)
 sum7(A) = mapreduce(x->x, plus, A)
 sum8(itr) = invoke(mapreduce, Tuple{Function, Function, Any}, x->x, plus, itr)
 sum9(A) = mapreduce(x->x, plus, A; init=0)
 sum10(itr) = invoke(Core.kwcall, Tuple{NamedTuple{(:init,),Tuple{Int}}, typeof(mapreduce), Function, Function, Any}, (init=0,), mapreduce, x->x, plus, itr)
-for f in (sum2, sum5, sum6, sum9, sum10)
+sum11(itr) = invoke(mapreduce, Tuple{Function, Function, Any}, init=0, x->x, plus, itr)
+for f in (sum2, sum5, sum6, sum61, sum9, sum10, sum11)
     @test sum(z) == f(z)
     @test sum(Int[]) == f(Int[]) == 0
     @test sum(Int[7]) == f(Int[7]) == 7