diff --git a/cpp/libsql/OPLibsqlBridge.cpp b/cpp/libsql/OPLibsqlBridge.cpp
index 9a5b374c..422b0938 100644
--- a/cpp/libsql/OPLibsqlBridge.cpp
+++ b/cpp/libsql/OPLibsqlBridge.cpp
@@ -750,27 +750,23 @@ opsqlite_libsql_execute_batch(DB const &db,
     throw std::runtime_error("No SQL commands provided");
   }
 
-  try {
-    int affectedRows = 0;
-    // opsqlite_libsql_execute(db, "BEGIN EXCLUSIVE TRANSACTION", nullptr);
-    for (int i = 0; i < commandCount; i++) {
-      auto command = commands->at(i);
-      // We do not provide a datastructure to receive query data because
-      // we don't need/want to handle this results in a batch execution
-      auto result = opsqlite_libsql_execute(db, command.sql, &command.params);
-      affectedRows += result.affectedRows;
-    }
-    // opsqlite_libsql_execute(db, "COMMIT", nullptr);
-    return BatchResult{
-        .affectedRows = affectedRows,
-        .commands = static_cast<int>(commandCount),
-    };
-  } catch (std::exception &exc) {
-    // opsqlite_libsql_execute(db, "ROLLBACK", nullptr);
-    return BatchResult{
-        .message = exc.what(),
-    };
-  }
+  int affectedRows = 0;
+  // Transaction control (BEGIN/COMMIT/ROLLBACK) is left to the JS side, so
+  // any exception here must propagate to reject the JS promise instead of
+  // being swallowed - otherwise the wrapping COMMIT would persist a partial
+  // batch instead of the ROLLBACK the caller expects.
+  for (int i = 0; i < commandCount; i++) {
+    auto command = commands->at(i);
+    // We do not provide a datastructure to receive query data because
+    // we don't need/want to handle this results in a batch execution
+    auto result = opsqlite_libsql_execute(db, command.sql, &command.params);
+    affectedRows += result.affectedRows;
+  }
+
+  return BatchResult{
+      .affectedRows = affectedRows,
+      .commands = static_cast<int>(commandCount),
+  };
 }
 
 } // namespace opsqlite
diff --git a/docs/docs/api.md b/docs/docs/api.md
index e4a680d2..57c27ca5 100644
--- a/docs/docs/api.md
+++ b/docs/docs/api.md
@@ -291,6 +291,14 @@ const res = await db.executeBatch(commands);
 console.log(`Batch affected ${result.rowsAffected} rows`);
 ```
 
+`executeBatch` runs the `BEGIN`/`COMMIT`/`ROLLBACK` statements that wrap the batch asynchronously, off the JS thread. For very large batches the `COMMIT` is where SQLite actually writes the WAL frames/fsyncs, so this keeps the JS thread free while that happens.
+
+If you need those transaction boundaries to run synchronously on the JS thread instead, use `executeBatchSync`, which has the same signature and behavior otherwise:
+
+```tsx
+const res = await db.executeBatchSync(commands);
+```
+
 In some scenarios, dynamic applications may need to get some metadata information about the returned result set.
 
 ## Blob support
diff --git a/example/src/tests/queries.ts b/example/src/tests/queries.ts
index 32798e89..b21a1515 100644
--- a/example/src/tests/queries.ts
+++ b/example/src/tests/queries.ts
@@ -483,6 +483,96 @@ describe("Queries tests", () => {
     ]);
   });
 
+  it("executeBatchSync", async () => {
+    const id1 = chance.integer();
+    const name1 = chance.name();
+    const age1 = chance.integer();
+    const networth1 = chance.floating();
+
+    const id2 = chance.integer();
+    const name2 = chance.name();
+    const age2 = chance.integer();
+    const networth2 = chance.floating();
+
+    const commands: SQLBatchTuple[] = [
+      ['SELECT * FROM "User"', []],
+      ['SELECT * FROM "User"'],
+      [
+        'INSERT INTO "User" (id, name, age, networth) VALUES(?, ?, ?, ?)',
+        [id1, name1, age1, networth1],
+      ],
+      [
+        'INSERT INTO "User" (id, name, age, networth) VALUES(?, ?, ?, ?)',
+        [[id2, name2, age2, networth2]],
+      ],
+    ];
+
+    await db.executeBatchSync(commands);
+
+    const res = await db.execute("SELECT * FROM User");
+
+    expect(res.rows).toDeepEqual([
+      { id: id1, name: name1, age: age1, networth: networth1, nickname: null },
+      {
+        id: id2,
+        name: name2,
+        age: age2,
+        networth: networth2,
+        nickname: null,
+      },
+    ]);
+  });
+
+  it("executeBatch rolls back on error", async () => {
+    const id1 = chance.integer();
+    const name1 = chance.name();
+    const age1 = chance.integer();
+    const networth1 = chance.floating();
+
+    const commands: SQLBatchTuple[] = [
+      [
+        'INSERT INTO "User" (id, name, age, networth) VALUES(?, ?, ?, ?)',
+        [id1, name1, age1, networth1],
+      ],
+      ["INSERT INTO [tableThatDoesNotExist] (id) VALUES(1)"],
+    ];
+
+    try {
+      await db.executeBatch(commands);
+      throw new Error("Should not resolve");
+    } catch (e) {
+      expect(((e as Error)?.message?.length ?? 0) > 0).toBe(true);
+    }
+
+    const res = await db.execute("SELECT * FROM User");
+    expect(res.rows).toDeepEqual([]);
+  });
+
+  it("executeBatchSync rolls back on error", async () => {
+    const id1 = chance.integer();
+    const name1 = chance.name();
+    const age1 = chance.integer();
+    const networth1 = chance.floating();
+
+    const commands: SQLBatchTuple[] = [
+      [
+        'INSERT INTO "User" (id, name, age, networth) VALUES(?, ?, ?, ?)',
+        [id1, name1, age1, networth1],
+      ],
+      ["INSERT INTO [tableThatDoesNotExist] (id) VALUES(1)"],
+    ];
+
+    try {
+      await db.executeBatchSync(commands);
+      throw new Error("Should not resolve");
+    } catch (e) {
+      expect(((e as Error)?.message?.length ?? 0) > 0).toBe(true);
+    }
+
+    const res = await db.execute("SELECT * FROM User");
+    expect(res.rows).toDeepEqual([]);
+  });
+
   it("Batch execute with BLOB", async () => {
     const db = open({
       name: "queries.sqlite",
diff --git a/src/functions.ts b/src/functions.ts
index f712b94a..ab6dd440 100644
--- a/src/functions.ts
+++ b/src/functions.ts
@@ -91,6 +91,39 @@ function enhanceDB(db: _InternalDB, options: DBParams): DB {
     },
     flushPendingReactiveQueries: db.flushPendingReactiveQueries,
     executeBatch: async (commands: SQLBatchTuple[]): Promise<BatchQueryResult> => {
+      async function run() {
+        try {
+          await enhancedDb.execute("BEGIN TRANSACTION;");
+
+          const res = await db.executeBatch(commands as any[]);
+
+          await enhancedDb.execute("COMMIT;");
+
+          await db.flushPendingReactiveQueries();
+
+          return res;
+        } catch (executionError) {
+          await enhancedDb.execute("ROLLBACK;");
+
+          throw executionError;
+        } finally {
+          lock.inProgress = false;
+          startNextTransaction();
+        }
+      }
+
+      return await new Promise((resolve, reject) => {
+        const tx: _PendingTransaction = {
+          start: () => {
+            run().then(resolve).catch(reject);
+          },
+        };
+
+        lock.queue.push(tx);
+        startNextTransaction();
+      });
+    },
+    executeBatchSync: async (commands: SQLBatchTuple[]): Promise<BatchQueryResult> => {
       async function run() {
         try {
           enhancedDb.executeSync("BEGIN TRANSACTION;");
diff --git a/src/functions.web.ts b/src/functions.web.ts
index f0f4867d..d464840b 100644
--- a/src/functions.web.ts
+++ b/src/functions.web.ts
@@ -300,6 +300,10 @@ function enhanceWebDb(db: _InternalDB, options: { name?: string; location?: stri
         rowsAffected: 0,
       };
     },
+    // Web has no synchronous native APIs, so there is no distinct blocking behavior to offer.
+    executeBatchSync: async (commands: SQLBatchTuple[]): Promise<BatchQueryResult> => {
+      return enhancedDb.executeBatch(commands);
+    },
     loadFile: async (_location: string): Promise<FileLoadResult> => {
       throw new Error("[op-sqlite] loadFile() is not supported on web.");
     },
diff --git a/src/types.ts b/src/types.ts
index 74c345d5..ea5484c9 100644
--- a/src/types.ts
+++ b/src/types.ts
@@ -248,10 +248,24 @@ export type DB = {
    * Executes all the queries in the params inside a single transaction
    *
    * It's faster than executing single queries as data is sent to the native side only once
+   *
+   * The BEGIN/COMMIT/ROLLBACK statements that wrap the batch are executed asynchronously,
+   * off the JS thread. Use this over `executeBatchSync` unless you specifically need the
+   * transaction boundaries to block the JS thread.
    * @param commands
    * @returns Promise<BatchQueryResult>
    */
   executeBatch: (commands: SQLBatchTuple[]) => Promise<BatchQueryResult>;
+  /**
+   * Same as `executeBatch` but the BEGIN/COMMIT/ROLLBACK statements that wrap the batch
+   * are executed synchronously on the JS thread. For large batches this can block the JS
+   * thread for a noticeable amount of time (the COMMIT is where SQLite writes the WAL
+   * frames/fsyncs), so prefer `executeBatch` unless you have a specific reason to need
+   * synchronous transaction boundaries.
+   * @param commands
+   * @returns Promise<BatchQueryResult>
+   */
+  executeBatchSync: (commands: SQLBatchTuple[]) => Promise<BatchQueryResult>;
   /**
    * Loads a SQLite Dump from disk. It will be the fastest way to execute a large set of queries as no JS is involved
    */