diff --git a/internal/engine/sqlite/parse.go b/internal/engine/sqlite/parse.go
index 13425b156e..195f98f795 100644
--- a/internal/engine/sqlite/parse.go
+++ b/internal/engine/sqlite/parse.go
@@ -37,12 +37,29 @@ func NewParser() *Parser {
 type Parser struct {
 }
 
+// runeToByteOffsets returns a slice mapping rune index i to the byte offset of
+// the i-th rune in s. A sentinel element equal to len(s) is appended so that
+// element numRunes is valid and equals the total byte length of s.
+//
+// This is needed because ANTLR4 stores the input as []rune and all token
+// positions (GetStart/GetStop) are rune indices, while Go string slicing is
+// byte-based.
+func runeToByteOffsets(s string) []int {
+	offsets := make([]int, 0, len(s)+1)
+	for i := range s {
+		offsets = append(offsets, i)
+	}
+	offsets = append(offsets, len(s))
+	return offsets
+}
+
 func (p *Parser) Parse(r io.Reader) ([]ast.Statement, error) {
 	blob, err := io.ReadAll(r)
 	if err != nil {
 		return nil, err
 	}
-	input := antlr.NewInputStream(string(blob))
+	src := string(blob)
+	input := antlr.NewInputStream(src)
 	lexer := parser.NewSQLiteLexer(input)
 	stream := antlr.NewCommonTokenStream(lexer, 0)
 	pp := parser.NewSQLiteParser(stream)
@@ -57,13 +74,17 @@ func (p *Parser) Parse(r io.Reader) ([]ast.Statement, error) {
 	if !ok {
 		return nil, fmt.Errorf("expected ParserContext; got %T\n", tree)
 	}
+	// ANTLR uses rune-based positions. Build a mapping from rune index to byte
+	// offset so we can store byte-based StmtLocation/StmtLen, which is what
+	// source.Pluck and the rest of the pipeline expect.
+	runeByteMap := runeToByteOffsets(src)
 	var stmts []ast.Statement
 	for _, istmt := range pctx.AllSql_stmt_list() {
 		list, ok := istmt.(*parser.Sql_stmt_listContext)
 		if !ok {
 			return nil, fmt.Errorf("expected Sql_stmt_listContext; got %T\n", istmt)
 		}
-		loc := 0
+		loc := 0 // rune offset of the current statement's start
 
 		for _, stmt := range list.AllSql_stmt() {
 			converter := &cc{}
@@ -72,12 +93,14 @@ func (p *Parser) Parse(r io.Reader) ([]ast.Statement, error) {
 				loc = stmt.GetStop().GetStop() + 2
 				continue
 			}
-			len := (stmt.GetStop().GetStop() + 1) - loc
+			runeEnd := stmt.GetStop().GetStop() + 1
+			byteStart := runeByteMap[loc]
+			byteEnd := runeByteMap[runeEnd]
 			stmts = append(stmts, ast.Statement{
 				Raw: &ast.RawStmt{
 					Stmt:         out,
-					StmtLocation: loc,
-					StmtLen:      len,
+					StmtLocation: byteStart,
+					StmtLen:      byteEnd - byteStart,
 				},
 			})
 			loc = stmt.GetStop().GetStop() + 2
diff --git a/internal/engine/sqlite/parse_test.go b/internal/engine/sqlite/parse_test.go
new file mode 100644
index 0000000000..71f124fdd7
--- /dev/null
+++ b/internal/engine/sqlite/parse_test.go
@@ -0,0 +1,97 @@
+package sqlite
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/sqlc-dev/sqlc/internal/source"
+)
+
+// TestParseNonASCIIComment verifies that non-ASCII characters in SQL comments
+// do not corrupt the plucked query text.
+//
+// ANTLR4 stores the input as []rune so all token positions are rune indices,
+// not byte offsets. source.Pluck (and the rest of the pipeline) treats
+// StmtLocation/StmtLen as byte offsets. For multi-byte UTF-8 characters the
+// two differ, which previously caused the plucked query to be truncated by one
+// byte per extra byte in each non-ASCII character.
+func TestParseNonASCIIComment(t *testing.T) {
+	p := NewParser()
+
+	tests := []struct {
+		name string
+		sql  string
+	}{
+		{
+			name: "2-byte char (U+00DC Ü) in dash comment",
+			sql:  "-- name: GetUser :one\n-- Ünïcode comment\nSELECT id FROM users WHERE id = ?",
+		},
+		{
+			name: "3-byte char (U+2665 ♥) in dash comment",
+			sql:  "-- name: GetUser :one\n-- ♥ love\nSELECT id FROM users WHERE id = ?",
+		},
+		{
+			name: "4-byte char (U+1D11E 𝄞) in dash comment",
+			sql:  "-- name: GetUser :one\n-- 𝄞 music\nSELECT id FROM users WHERE id = ?",
+		},
+		{
+			name: "multiple non-ASCII chars in comment",
+			sql:  "-- name: GetUser :one\n-- héllo wörld\nSELECT id FROM users WHERE id = ?",
+		},
+		{
+			name: "non-ASCII only in first of two statements",
+			sql:  "-- name: Q1 :one\n-- Ü\nSELECT 1;\n\n-- name: Q2 :one\nSELECT 2",
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			stmts, err := p.Parse(strings.NewReader(tc.sql))
+			if err != nil {
+				t.Fatalf("Parse error: %v", err)
+			}
+			if len(stmts) == 0 {
+				t.Fatal("expected at least one statement")
+			}
+
+			// For every parsed statement, verify that the plucked text is a
+			// valid substring of the original SQL (not truncated mid-character).
+			for i, stmt := range stmts {
+				raw := stmt.Raw
+				plucked, err := source.Pluck(tc.sql, raw.StmtLocation, raw.StmtLen)
+				if err != nil {
+					t.Fatalf("stmt %d: Pluck error: %v", i, err)
+				}
+				if !strings.Contains(tc.sql, plucked) {
+					t.Errorf("stmt %d: plucked text is not a substring of the input\ngot:  %q\ninput: %q", i, plucked, tc.sql)
+				}
+				if plucked == "" {
+					t.Errorf("stmt %d: plucked text is empty", i)
+				}
+			}
+
+			// For the single-statement cases the plucked text must equal the
+			// full input, since there is exactly one statement and no trailing
+			// semicolon to exclude.
+			if len(stmts) == 1 {
+				raw := stmts[0].Raw
+				plucked, _ := source.Pluck(tc.sql, raw.StmtLocation, raw.StmtLen)
+				if plucked != tc.sql {
+					t.Errorf("single-statement pluck mismatch\ngot:  %q\nwant: %q", plucked, tc.sql)
+				}
+			}
+
+			// For the two-statement case, verify each statement contains its
+			// expected SELECT.
+			if len(stmts) == 2 {
+				for i, want := range []string{"SELECT 1", "SELECT 2"} {
+					raw := stmts[i].Raw
+					plucked, _ := source.Pluck(tc.sql, raw.StmtLocation, raw.StmtLen)
+					if !strings.Contains(plucked, want) {
+						t.Errorf("stmt %d: plucked text %q does not contain %q", i, plucked, want)
+					}
+				}
+			}
+		})
+	}
+}