diff --git a/Cargo.toml b/Cargo.toml
index 659b4817..19065695 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -82,6 +82,7 @@ parking_lot = "0.12.1"
 pollster = "0.3.0"
 rayon = "1.7.0"
 tokio = { version = "1.28.1", features = ["full"] }
+strum = { version = "0.25.0", features = ["derive"] }
 
 # data structure and algorithm
 append-only-vec = "0.1.2"
diff --git a/compiler/Cargo.toml b/compiler/Cargo.toml
index 776d29aa..6b82a8e2 100644
--- a/compiler/Cargo.toml
+++ b/compiler/Cargo.toml
@@ -23,7 +23,7 @@ sha2.workspace = true
 flate2.workspace = true
 ecow.workspace = true
 instant.workspace = true
-strum = { version = "0.25.0", features = ["derive"] }
+strum.workspace = true
 
 serde.workspace = true
 serde_json.workspace = true
diff --git a/compiler/src/lib.rs b/compiler/src/lib.rs
index 3611783d..4b25cd14 100644
--- a/compiler/src/lib.rs
+++ b/compiler/src/lib.rs
@@ -41,7 +41,7 @@ pub mod vfs;
 pub mod world;
 
 /// Diff and parse the source code.
-mod parser;
+pub mod parser;
 mod utils;
 
 /// Convenient services over [`world::CompilerWorld`].
diff --git a/compiler/src/parser/mod.rs b/compiler/src/parser/mod.rs
index f04cf259..9e0dfd3b 100644
--- a/compiler/src/parser/mod.rs
+++ b/compiler/src/parser/mod.rs
@@ -8,7 +8,8 @@ use typst::{diag::FileResult, syntax::Source};
 use typst_ts_core::TypstFileId;
 
 pub use semantic_tokens::{
-    get_semantic_tokens_full, get_semantic_tokens_legend, SemanticToken, SemanticTokensLegend,
+    get_semantic_tokens_full, get_semantic_tokens_legend, OffsetEncoding, SemanticToken,
+    SemanticTokensLegend,
 };
 
 pub fn reparse(source_id: TypstFileId, prev: Option<Source>, next: String) -> FileResult<Source> {
diff --git a/compiler/src/parser/semantic_tokens.rs b/compiler/src/parser/semantic_tokens.rs
index 0601e4ec..38596b49 100644
--- a/compiler/src/parser/semantic_tokens.rs
+++ b/compiler/src/parser/semantic_tokens.rs
@@ -33,12 +33,26 @@ pub fn get_semantic_tokens_legend() -> SemanticTokensLegend {
     }
 }
 
-pub fn get_semantic_tokens_full(source: &Source) -> Vec<SemanticToken> {
+pub enum OffsetEncoding {
+    Utf8,
+    Utf16,
+}
+
+pub fn get_semantic_tokens_full(source: &Source, encoding: OffsetEncoding) -> Vec<SemanticToken> {
     let root = LinkedNode::new(source.root());
-    let mut full = tokenize_tree(source, &root, ModifierSet::empty());
+    let mut full = tokenize_tree(&root, ModifierSet::empty());
 
     let mut init = (0, 0);
     for token in full.iter_mut() {
+        // resolve offset to position
+        let offset = ((token.delta_line as u64) << 32) | token.delta_start_character as u64;
+        let position = (match encoding {
+            OffsetEncoding::Utf8 => offset_to_position_utf8,
+            OffsetEncoding::Utf16 => offset_to_position_utf16,
+        })(offset as usize, source);
+        token.delta_line = position.0;
+        token.delta_start_character = position.1;
+
         let next = (token.delta_line, token.delta_start_character);
         token.delta_line -= init.0;
         if token.delta_line == 0 {
@@ -50,30 +64,22 @@ pub fn get_semantic_tokens_full(source: &Source) -> Vec<SemanticToken> {
     full
 }
 
-fn tokenize_single_node(
-    ctx: &Source,
-    node: &LinkedNode,
-    modifiers: ModifierSet,
-) -> Option<SemanticToken> {
+fn tokenize_single_node(node: &LinkedNode, modifiers: ModifierSet) -> Option<SemanticToken> {
     let is_leaf = node.children().next().is_none();
 
     token_from_node(node)
         .or_else(|| is_leaf.then_some(TokenType::Text))
-        .map(|token_type| SemanticToken::new(ctx, token_type, modifiers, node))
+        .map(|token_type| SemanticToken::new(token_type, modifiers, node))
 }
 
 /// Tokenize a node and its children
-fn tokenize_tree(
-    ctx: &Source,
-    root: &LinkedNode<'_>,
-    parent_modifiers: ModifierSet,
-) -> Vec<SemanticToken> {
+fn tokenize_tree(root: &LinkedNode<'_>, parent_modifiers: ModifierSet) -> Vec<SemanticToken> {
     let modifiers = parent_modifiers | modifiers_from_node(root);
 
-    let token = tokenize_single_node(ctx, root, modifiers).into_iter();
+    let token = tokenize_single_node(root, modifiers).into_iter();
     let children = root
         .children()
-        .flat_map(move |child| tokenize_tree(ctx, &child, modifiers));
+        .flat_map(move |child| tokenize_tree(&child, modifiers));
     token.chain(children).collect()
 }
 
@@ -86,21 +92,17 @@ pub struct SemanticToken {
 }
 
 impl SemanticToken {
-    pub fn new(
-        ctx: &Source,
-        token_type: TokenType,
-        modifiers: ModifierSet,
-        node: &LinkedNode,
-    ) -> Self {
+    fn new(token_type: TokenType, modifiers: ModifierSet, node: &LinkedNode) -> Self {
         let source = node.get().clone().into_text();
 
-        let position = offset_to_position(node.offset(), ctx);
+        let raw_position = node.offset() as u64;
+        let raw_position = ((raw_position >> 32) as u32, raw_position as u32);
 
         Self {
             token_type: token_type as u32,
             token_modifiers: modifiers.bitset(),
-            delta_line: position.0,
-            delta_start_character: position.1,
+            delta_line: raw_position.0,
+            delta_start_character: raw_position.1,
             length: source.chars().map(char::len_utf16).sum::<usize>() as u32,
         }
     }
@@ -202,7 +204,14 @@ fn token_from_hashtag(hashtag: &LinkedNode) -> Option<TokenType> {
         .and_then(token_from_node)
 }
 
-fn offset_to_position(typst_offset: usize, typst_source: &Source) -> (u32, u32) {
+fn offset_to_position_utf8(typst_offset: usize, typst_source: &Source) -> (u32, u32) {
+    let line_index = typst_source.byte_to_line(typst_offset).unwrap();
+    let column_index = typst_source.byte_to_column(typst_offset).unwrap();
+
+    (line_index as u32, column_index as u32)
+}
+
+fn offset_to_position_utf16(typst_offset: usize, typst_source: &Source) -> (u32, u32) {
     let line_index = typst_source.byte_to_line(typst_offset).unwrap();
 
     let lsp_line = line_index as u32;
diff --git a/compiler/src/world.rs b/compiler/src/world.rs
index be0de5eb..ae6f1f38 100644
--- a/compiler/src/world.rs
+++ b/compiler/src/world.rs
@@ -24,7 +24,8 @@ use crate::{
     dependency::{DependencyTree, DependentFileInfo},
     package::Registry as PackageRegistry,
     parser::{
-        get_semantic_tokens_full, get_semantic_tokens_legend, SemanticToken, SemanticTokensLegend,
+        get_semantic_tokens_full, get_semantic_tokens_legend, OffsetEncoding, SemanticToken,
+        SemanticTokensLegend,
     },
     service::WorkspaceProvider,
     time::SystemTime,
@@ -207,17 +208,21 @@ impl<F: CompilerFeat> CompilerWorld<F> {
         Arc::new(get_semantic_tokens_legend())
     }
 
-    pub fn get_semantic_tokens(&self, file_path: Option<String>) -> Arc<Vec<SemanticToken>> {
-        Arc::new(get_semantic_tokens_full(
-            &file_path
-                .and_then(|e| {
-                    let relative_path = Path::new(&e).strip_prefix(&self.workspace_root()).ok()?;
+    pub fn get_semantic_tokens(
+        &self,
+        file_path: Option<String>,
+        encoding: OffsetEncoding,
+    ) -> Arc<Vec<SemanticToken>> {
+        let src = &file_path
+            .and_then(|e| {
+                let relative_path = Path::new(&e).strip_prefix(&self.workspace_root()).ok()?;
+
+                let source_id = FileId::new(None, VirtualPath::new(relative_path));
+                self.source(source_id).ok()
+            })
+            .unwrap_or_else(|| self.main());
 
-                    let source_id = FileId::new(None, VirtualPath::new(relative_path));
-                    self.source(source_id).ok()
-                })
-                .unwrap_or_else(|| self.main()),
-        ))
+        Arc::new(get_semantic_tokens_full(src, encoding))
     }
 
     fn map_source_or_default<T>(
diff --git a/packages/compiler/src/lib.rs b/packages/compiler/src/lib.rs
index 51624480..de36a295 100644
--- a/packages/compiler/src/lib.rs
+++ b/packages/compiler/src/lib.rs
@@ -7,6 +7,7 @@ pub use typst_ts_compiler::*;
 use typst_ts_compiler::{
     font::web::BrowserFontSearcher,
     package::browser::ProxyRegistry,
+    parser::OffsetEncoding,
     service::{CompileDriverImpl, Compiler},
     vfs::browser::ProxyAccessModel,
     world::WorldSnapshot,
@@ -206,6 +207,7 @@ impl TypstCompiler {
 
     pub fn get_semantic_tokens(
         &mut self,
+        offset_encoding: String,
         file_path: Option<String>,
         result_id: Option<String>,
     ) -> Result<js_sys::Object, JsValue> {
@@ -215,7 +217,16 @@ impl TypstCompiler {
             );
         }
 
-        let tokens = self.compiler.world_mut().get_semantic_tokens(file_path);
+        let tokens = self.compiler.world_mut().get_semantic_tokens(
+            file_path,
+            match offset_encoding.as_str() {
+               "utf-16" => OffsetEncoding::Utf16,
+              "utf-8" => OffsetEncoding::Utf8,
+                _ => {
+                    return Err(error_once!("Unsupported offset encoding", offset_encoding: offset_encoding).into());
+                }
+            },
+        );
         let mut result = Vec::new();
         for token in tokens.iter() {
             result.push(token.delta_line);
diff --git a/packages/typst.ts/src/compiler.mts b/packages/typst.ts/src/compiler.mts
index 2758adb3..e867147c 100644
--- a/packages/typst.ts/src/compiler.mts
+++ b/packages/typst.ts/src/compiler.mts
@@ -116,8 +116,20 @@ export interface TypstCompiler {
   /**
    * experimental
    * See Semantic tokens: https://github.com/microsoft/vscode/issues/86415
+   *
+   * @param {string} opts.mainFilePath - The path of the main file.
+   * @param {string} opts.resultId - The id of the result.
+   * @param {string} opts.offsetEncoding - The encoding of the offset.
+   *   - 'utf-16': the offset is encoded in utf-16.
+   *   - 'utf-8': the offset is encoded in utf-8.
+   *   @default 'utf-16'
+   * @returns {Promise<SemanticTokens>} - The semantic tokens.
    */
-  getSemanticTokens(opts: { mainFilePath: string; resultId?: string }): Promise<SemanticTokens>;
+  getSemanticTokens(opts: {
+    mainFilePath: string;
+    resultId?: string;
+    offsetEncoding?: string;
+  }): Promise<SemanticTokens>;
 }
 
 const gCompilerModule = new LazyWasmModule(async (bin?: any) => {
@@ -193,10 +205,20 @@ class TypstCompilerDriver {
     });
   }
 
-  getSemanticTokens(opts: { mainFilePath: string; resultId?: string }): Promise<SemanticTokens> {
+  getSemanticTokens(opts: {
+    mainFilePath: string;
+    resultId?: string;
+    offsetEncoding?: string;
+  }): Promise<SemanticTokens> {
     return new Promise<SemanticTokens>(resolve => {
       this.compiler.reset();
-      resolve(this.compiler.get_semantic_tokens(opts.mainFilePath, opts.resultId) as any);
+      resolve(
+        this.compiler.get_semantic_tokens(
+          opts.offsetEncoding || 'utf-16',
+          opts.mainFilePath,
+          opts.resultId,
+        ) as any,
+      );
     });
   }
 
diff --git a/packages/typst.ts/src/contrib/snippet.mts b/packages/typst.ts/src/contrib/snippet.mts
index 3f052fbb..fdf2e5e5 100644
--- a/packages/typst.ts/src/contrib/snippet.mts
+++ b/packages/typst.ts/src/contrib/snippet.mts
@@ -434,6 +434,8 @@ export class TypstSnippet {
 
   /**
    * Get semantic tokens for the document.
+   * See {@link SweetCompileOptions}.
+   * See {@link TypstCompiler#getSemanticTokens}.
    */
   async getSemanticTokens(
     opts: SweetCompileOptions & { resultId?: string },