diff --git a/Cargo.lock b/Cargo.lock
index ca49a699d1..e5efe24226 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -159,9 +159,9 @@ dependencies = [
 
 [[package]]
 name = "async-openai"
-version = "0.34.0"
+version = "0.40.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ec08254d61379df136135d3d1ac04301be7699fd7d9e57655c63ac7d650a6922"
+checksum = "01fa0605961e005565ddbeff2d8f9148b8893128381ee3bec3a5df4e6a8d0f74"
 dependencies = [
  "derive_builder",
  "getrandom 0.3.4",
diff --git a/Cargo.toml b/Cargo.toml
index a8080c8529..f658fecb6f 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -135,7 +135,7 @@ gray_matter = "0.3.2"
 num-format = "0.4"
 humantime = "2.1.0"
 dashmap = "7.0.0-rc2"
-async-openai = { version = "0.34.0", default-features = false, features = ["response-types"] } # Using only types, not the API client - reduces dependencies
+async-openai = { version = "0.40.0", default-features = false, features = ["response-types"] } # Using only types, not the API client - reduces dependencies
 gix = "0.83"
 google-cloud-auth = "1.8.0" # Google Cloud authentication with automatic token refresh