diff --git a/crates/language_models/src/provider/open_ai_compatible.rs b/crates/language_models/src/provider/open_ai_compatible.rs
index c1f9f70a15..da165a77b5 100644
--- a/crates/language_models/src/provider/open_ai_compatible.rs
+++ b/crates/language_models/src/provider/open_ai_compatible.rs
@@ -322,7 +322,7 @@ impl LanguageModel for OpenAiCompatibleLanguageModel {
 
     fn stream_completion(
         &self,
-        request: LanguageModelRequest,
+        mut request: LanguageModelRequest,
         cx: &AsyncApp,
     ) -> BoxFuture<
         'static,
@@ -334,6 +334,12 @@ impl LanguageModel for OpenAiCompatibleLanguageModel {
             LanguageModelCompletionError,
         >,
     > {
+        // `speed` can leak in from a parent thread's model; this provider never
+        // supports fast mode, and arbitrary compatible endpoints reject `service_tier`.
+        if !self.supports_fast_mode() {
+            request.speed = None;
+        }
+
         if self.model.capabilities.chat_completions {
             let request = into_open_ai(
                 request,