diff --git a/stdlib/REPL/src/LineEdit.jl b/stdlib/REPL/src/LineEdit.jl
index 01daef9d66899..01508d50b22fb 100644
--- a/stdlib/REPL/src/LineEdit.jl
+++ b/stdlib/REPL/src/LineEdit.jl
@@ -381,7 +381,16 @@ function check_for_hint(s::MIState)
         # Requires making space for them earlier in refresh_multi_line
         return clear_hint(st)
     end
-    completions, partial, should_complete = complete_line(st.p.complete, st, s.active_module)::Tuple{Vector{String},String,Bool}
+    # spawn a new task for hint completion to avoid blocking the main thread
+    # if the completion task doesn't finish quickly, abandon hinting and return early
+    completion_task = Threads.@spawn :default complete_line(st.p.complete, st, s.active_module)
+    done_quickly = false
+    completion_timer = Timer(0.01) do _
+        done_quickly = istaskdone(completion_task)
+    end
+    wait(completion_timer); close(completion_timer)
+    done_quickly || return clear_hint(st) # TODO kill `completion_task`?
+    completions, partial, should_complete = fetch(completion_task)::Tuple{Vector{String},String,Bool}
     isempty(completions) && return clear_hint(st)
     # Don't complete for single chars, given e.g. `x` completes to `xor`
     if length(partial) > 1 && should_complete
@@ -2395,7 +2404,8 @@ end
 # jump_spaces: if cursor is on a ' ', move it to the first non-' ' char on the right
 # if `delete_trailing`, ignore trailing ' ' by deleting them
 function edit_tab(s::MIState, jump_spaces::Bool=false, delete_trailing::Bool=jump_spaces)
-    tab_should_complete(s) && return complete_line(s)
+    # tab complete on a separate thread (if available) to avoid blocking the main thread
+    tab_should_complete(s) && return Threads.@spawn :default complete_line(s)
     set_action!(s, :edit_insert_tab)
     push_undo(s)
     edit_insert_tab(buffer(s), jump_spaces, delete_trailing) || pop_undo(s)
@@ -2503,7 +2513,10 @@ AnyDict(
     "^_" => (s::MIState,o...)->edit_undo!(s),
     "\e_" => (s::MIState,o...)->edit_redo!(s),
     # Show hints at what tab complete would do by default
-    "*" => (s::MIState,data,c::StringLike)->(edit_insert(s, c); check_for_hint(s) && refresh_line(s)),
+    "*" => function (s::MIState,data,c::StringLike)
+        edit_insert(s, c)
+        check_for_hint(s) && refresh_line(s)
+    end,
     "^U" => (s::MIState,o...)->edit_kill_line_backwards(s),
     "^K" => (s::MIState,o...)->edit_kill_line_forwards(s),
     "^Y" => (s::MIState,o...)->edit_yank(s),
diff --git a/stdlib/REPL/src/REPLCompletions.jl b/stdlib/REPL/src/REPLCompletions.jl
index da5d20270c5b6..f1e58606b5185 100644
--- a/stdlib/REPL/src/REPLCompletions.jl
+++ b/stdlib/REPL/src/REPLCompletions.jl
@@ -646,6 +646,10 @@ function resolve_toplevel_symbols!(src::Core.CodeInfo, mod::Module)
     return src
 end
 
+# the lock to avoid concurrent inference from multiple completion tasks, which could happen
+# if user types fast rapidly (otherwise end up with corrupt inference states)
+const REPL_INFERENCE_LOCK = ReentrantLock()
+
 # lower `ex` and run type inference on the resulting top-level expression
 function repl_eval_ex(@nospecialize(ex), context_module::Module; limit_aggressive_inference::Bool=false)
     if (isexpr(ex, :toplevel) || isexpr(ex, :tuple)) && !isempty(ex.args)
@@ -662,6 +666,17 @@ function repl_eval_ex(@nospecialize(ex), context_module::Module; limit_aggressiv
     end
     lwr isa Expr || return Const(lwr) # `ex` is literal
     isexpr(lwr, :thunk) || return nothing # lowered to `Expr(:error, ...)` or similar
+    trylock(REPL_INFERENCE_LOCK) || return nothing
+    result = try
+        repl_infer_ex(lwr, context_module, limit_aggressive_inference)
+    finally
+        unlock(REPL_INFERENCE_LOCK)
+    end
+    result === Union{} && return nothing # for whatever reason, callers expect this as the Bottom and/or Top type instead
+    return result
+end
+
+function repl_infer_ex(lwr::Expr, context_module::Module, limit_aggressive_inference::Bool=false)
     src = lwr.args[1]::Core.CodeInfo
 
     # construct top-level `MethodInstance`
@@ -680,9 +695,7 @@ function repl_eval_ex(@nospecialize(ex), context_module::Module; limit_aggressiv
     #      potential invalidations of `Core.Compiler` methods.
     Base.invoke_in_world(COMPLETION_WORLD[], CC.typeinf, interp, frame)
 
-    result = frame.result.result
-    result === Union{} && return nothing # for whatever reason, callers expect this as the Bottom and/or Top type instead
-    return result
+    return frame.result.result
 end
 
 # `COMPLETION_WORLD[]` will be initialized within `__init__`