diff --git a/bench/bench.zig b/bench/bench.zig
index 83afc8b4..a290a7b9 100644
--- a/bench/bench.zig
+++ b/bench/bench.zig
@@ -1,5 +1,6 @@
 const std = @import("std");
 const vaxis = @import("vaxis");
+const uucode = @import("uucode");
 
 fn parseIterations(allocator: std.mem.Allocator) !usize {
     var args = try std.process.argsWithAllocator(allocator);
@@ -20,6 +21,103 @@ fn printResults(writer: anytype, label: []const u8, iterations: usize, elapsed_n
     );
 }
 
+// Mirrors the pre-fast-path parseGround work for ASCII to provide a baseline.
+fn parseAsciiSlow(input: []const u8) !vaxis.Parser.Result {
+    std.debug.assert(input.len > 0);
+    var iter = uucode.utf8.Iterator.init(input);
+    const first_cp = iter.next() orelse return error.InvalidUTF8;
+
+    var n: usize = std.unicode.utf8CodepointSequenceLength(first_cp) catch return error.InvalidUTF8;
+
+    var code = first_cp;
+    var grapheme_iter = uucode.grapheme.Iterator(uucode.utf8.Iterator).init(.init(input));
+    var grapheme_len: usize = 0;
+    var cp_count: usize = 0;
+
+    while (grapheme_iter.next()) |result| {
+        cp_count += 1;
+        if (result.is_break) {
+            grapheme_len = grapheme_iter.i;
+            break;
+        }
+    }
+
+    if (grapheme_len > 0) {
+        n = grapheme_len;
+        if (cp_count > 1) {
+            code = vaxis.Key.multicodepoint;
+        }
+    }
+
+    const key: vaxis.Key = .{ .codepoint = code, .text = input[0..n] };
+    return .{ .event = .{ .key_press = key }, .n = n };
+}
+
+fn benchParseFast(writer: anytype, label: []const u8, parser: *vaxis.Parser, input: []const u8, iterations: usize) !void {
+    var timer = try std.time.Timer.start();
+    var i: usize = 0;
+    while (i < iterations) : (i += 1) {
+        const result = try parser.parse(input, null);
+        std.mem.doNotOptimizeAway(result);
+    }
+    const elapsed_ns = timer.read();
+    try printResults(writer, label, iterations, elapsed_ns, input.len * iterations);
+}
+
+fn benchParseSlow(writer: anytype, label: []const u8, input: []const u8, iterations: usize) !void {
+    var timer = try std.time.Timer.start();
+    var i: usize = 0;
+    while (i < iterations) : (i += 1) {
+        const result = try parseAsciiSlow(input);
+        std.mem.doNotOptimizeAway(result);
+    }
+    const elapsed_ns = timer.read();
+    try printResults(writer, label, iterations, elapsed_ns, input.len * iterations);
+}
+
+fn benchParseStreamSlow(writer: anytype, label: []const u8, parser: *vaxis.Parser, input: []const u8, iterations: usize) !void {
+    var timer = try std.time.Timer.start();
+    var i: usize = 0;
+    while (i < iterations) : (i += 1) {
+        var idx: usize = 0;
+        while (idx < input.len) {
+            const next = if (idx + 1 < input.len) input[idx + 1] else null;
+            if (input[idx] >= 0x20 and input[idx] <= 0x7E and (next == null or next.? < 0x80)) {
+                const result = try parseAsciiSlow(input[idx..]);
+                if (result.n == 0) break;
+                idx += result.n;
+                std.mem.doNotOptimizeAway(result);
+                continue;
+            }
+
+            const result = try parser.parse(input[idx..], null);
+            if (result.n == 0) break;
+            idx += result.n;
+            std.mem.doNotOptimizeAway(result);
+        }
+        std.mem.doNotOptimizeAway(idx);
+    }
+    const elapsed_ns = timer.read();
+    try printResults(writer, label, iterations, elapsed_ns, input.len * iterations);
+}
+
+fn benchParseStreamFast(writer: anytype, label: []const u8, parser: *vaxis.Parser, input: []const u8, iterations: usize) !void {
+    var timer = try std.time.Timer.start();
+    var i: usize = 0;
+    while (i < iterations) : (i += 1) {
+        var idx: usize = 0;
+        while (idx < input.len) {
+            const result = try parser.parse(input[idx..], null);
+            if (result.n == 0) break;
+            idx += result.n;
+            std.mem.doNotOptimizeAway(result);
+        }
+        std.mem.doNotOptimizeAway(idx);
+    }
+    const elapsed_ns = timer.read();
+    try printResults(writer, label, iterations, elapsed_ns, input.len * iterations);
+}
+
 pub fn main() !void {
     var gpa = std.heap.GeneralPurposeAllocator(.{}){};
     defer _ = gpa.deinit();
@@ -59,4 +157,13 @@ pub fn main() !void {
     const dirty_ns = timer.read();
     const dirty_bytes: usize = dirty_writer.writer.end;
     try printResults(stdout, "dirty", iterations, dirty_ns, dirty_bytes);
+
+    var parser: vaxis.Parser = .{};
+    const ascii_input = "a";
+    try benchParseSlow(stdout, "parse_ground_ascii_slow", ascii_input, iterations);
+    try benchParseFast(stdout, "parse_ground_ascii_fast", &parser, ascii_input, iterations);
+
+    const mixed_input = "hello \x1b[A世界 1️⃣ 👩‍🚀!\r";
+    try benchParseStreamSlow(stdout, "parse_stream_mixed_slow", &parser, mixed_input, iterations);
+    try benchParseStreamFast(stdout, "parse_stream_mixed_fast", &parser, mixed_input, iterations);
 }
diff --git a/build.zig b/build.zig
index 2265ac5f..ede5bc7c 100644
--- a/build.zig
+++ b/build.zig
@@ -76,6 +76,7 @@ pub fn build(b: *std.Build) void {
             .optimize = optimize,
             .imports = &.{
                 .{ .name = "vaxis", .module = vaxis_mod },
+                .{ .name = "uucode", .module = uucode_dep.module("uucode") },
             },
         }),
     });
diff --git a/src/Parser.zig b/src/Parser.zig
index 6358269a..b4c3075b 100644
--- a/src/Parser.zig
+++ b/src/Parser.zig
@@ -83,6 +83,18 @@ inline fn parseGround(input: []const u8) !Result {
     std.debug.assert(input.len > 0);
 
     const b = input[0];
+    // Fast path for printable ASCII only (0x20..0x7E). Control bytes are
+    // mapped to special key handling and ESC can start control sequences.
+    // Require the next byte to be ASCII (or absent) to avoid multi-codepoint
+    // graphemes like combining marks/keycap sequences that start with ASCII
+    // but continue with non-ASCII bytes.
+    if (b >= 0x20 and b <= 0x7E and (input.len == 1 or input[1] < 0x80)) {
+        return .{
+            .event = .{ .key_press = .{ .codepoint = b, .text = input[0..1] } },
+            .n = 1,
+        };
+    }
+
     var n: usize = 1;
     // ground state generates keypresses when parsing input. We
     // generally get ascii characters, but anything less than