language_models: Support fast mode on ChatGPT subscription provider (#57436)

Same mechanism as for BYOK: `service_tier == priority`. Most of the work
is already done. When validating this in manual testing, I noticed we
get back `service_tier == auto` in the response, unlike in the regular
OpenAI API scenario with BYOK, but apparently [it doesn't mean priority
tier wasn't
applied](https://github.com/openai/codex/issues/14204#issuecomment-4033184620).
It's not a hard confirmation, but the model does seem to respond faster
when I toggle fast mode on.

Release Notes:

- Added Fast Mode (priority service tier) support to OpenAI models used
through the ChatGPT subscription provider.

crates/language_models/src/provider/openai_subscribed.rs

@@ -345,6 +345,13 @@ impl ChatGptModel {
     fn supports_prompt_cache_key(&self) -> bool {
         true
     }
+
+    fn supports_priority(&self) -> bool {
+        match self {
+            Self::Gpt55 | Self::Gpt54 => true,
+            Self::Gpt54Mini | Self::Gpt53Codex | Self::Gpt52 => false,
+        }
+    }
 }
 
 struct OpenAiSubscribedLanguageModel {
@@ -392,6 +399,10 @@ impl LanguageModel for OpenAiSubscribedLanguageModel {
         true
     }
 
+    fn supports_fast_mode(&self) -> bool {
+        self.model.supports_priority()
+    }
+
     fn supported_effort_levels(&self) -> Vec<LanguageModelEffortLevel> {
         let default_effort = self.model.default_reasoning_effort();
         self.model
@@ -431,7 +442,7 @@ impl LanguageModel for OpenAiSubscribedLanguageModel {
 
     fn stream_completion(
         &self,
-        request: LanguageModelRequest,
+        mut request: LanguageModelRequest,
         cx: &AsyncApp,
     ) -> BoxFuture<
         'static,
@@ -443,6 +454,10 @@ impl LanguageModel for OpenAiSubscribedLanguageModel {
             LanguageModelCompletionError,
         >,
     > {
+        if !self.model.supports_priority() {
+            request.speed = None;
+        }
+
         // The Codex backend rejects `max_output_tokens` (`Unsupported parameter`),
         // unlike the public OpenAI Responses API. Pass `None` so the field is
         // omitted from the serialized request body entirely.

crates/open_ai/src/responses.rs

@@ -335,6 +335,8 @@ pub struct ResponseSummary {
     pub usage: Option<ResponseUsage>,
     #[serde(default)]
     pub output: Vec<ResponseOutputItem>,
+    #[serde(default)]
+    pub service_tier: Option<crate::ServiceTier>,
 }
 
 #[derive(Deserialize, Debug, Default, Clone)]