From 2a57b160b03c8e8543fdae12a0c191ed1a985e54 Mon Sep 17 00:00:00 2001
From: Oleksiy Syvokon <oleksiy@zed.dev>
Date: Fri, 15 Aug 2025 13:54:24 +0300
Subject: [PATCH] openai: Don't send prompt_cache_key for OpenAI-compatible
 models (#36231)

Some APIs fail when they get this parameter

Closes #36215

Release Notes:

- Fixed OpenAI-compatible providers that don't support prompt caching
and/or reasoning
---
 crates/language_models/src/provider/cloud.rs              | 1 +
 crates/language_models/src/provider/open_ai.rs            | 8 +++++++-
 crates/language_models/src/provider/open_ai_compatible.rs | 5 ++++-
 crates/language_models/src/provider/vercel.rs             | 1 +
 crates/language_models/src/provider/x_ai.rs               | 1 +
 crates/open_ai/src/open_ai.rs                             | 7 +++++++
 crates/vercel/src/vercel.rs                               | 4 ++++
 crates/x_ai/src/x_ai.rs                                   | 4 ++++
 8 files changed, 29 insertions(+), 2 deletions(-)

diff --git a/crates/language_models/src/provider/cloud.rs b/crates/language_models/src/provider/cloud.rs
index ff8048040e..c1337399f9 100644
--- a/crates/language_models/src/provider/cloud.rs
+++ b/crates/language_models/src/provider/cloud.rs
@@ -941,6 +941,7 @@ impl LanguageModel for CloudLanguageModel {
                     request,
                     model.id(),
                     model.supports_parallel_tool_calls(),
+                    model.supports_prompt_cache_key(),
                     None,
                     None,
                 );
diff --git a/crates/language_models/src/provider/open_ai.rs b/crates/language_models/src/provider/open_ai.rs
index 725027b2a7..eaf8d885b3 100644
--- a/crates/language_models/src/provider/open_ai.rs
+++ b/crates/language_models/src/provider/open_ai.rs
@@ -370,6 +370,7 @@ impl LanguageModel for OpenAiLanguageModel {
             request,
             self.model.id(),
             self.model.supports_parallel_tool_calls(),
+            self.model.supports_prompt_cache_key(),
             self.max_output_tokens(),
             self.model.reasoning_effort(),
         );
@@ -386,6 +387,7 @@ pub fn into_open_ai(
     request: LanguageModelRequest,
     model_id: &str,
     supports_parallel_tool_calls: bool,
+    supports_prompt_cache_key: bool,
     max_output_tokens: Option<u64>,
     reasoning_effort: Option<ReasoningEffort>,
 ) -> open_ai::Request {
@@ -477,7 +479,11 @@ pub fn into_open_ai(
         } else {
             None
         },
-        prompt_cache_key: request.thread_id,
+        prompt_cache_key: if supports_prompt_cache_key {
+            request.thread_id
+        } else {
+            None
+        },
         tools: request
             .tools
             .into_iter()
diff --git a/crates/language_models/src/provider/open_ai_compatible.rs b/crates/language_models/src/provider/open_ai_compatible.rs
index 6e912765cd..5f546f5219 100644
--- a/crates/language_models/src/provider/open_ai_compatible.rs
+++ b/crates/language_models/src/provider/open_ai_compatible.rs
@@ -355,10 +355,13 @@ impl LanguageModel for OpenAiCompatibleLanguageModel {
             LanguageModelCompletionError,
         >,
     > {
+        let supports_parallel_tool_call = true;
+        let supports_prompt_cache_key = false;
         let request = into_open_ai(
             request,
             &self.model.name,
-            true,
+            supports_parallel_tool_call,
+            supports_prompt_cache_key,
             self.max_output_tokens(),
             None,
         );
diff --git a/crates/language_models/src/provider/vercel.rs b/crates/language_models/src/provider/vercel.rs
index 57a89ba4aa..9f447cb68b 100644
--- a/crates/language_models/src/provider/vercel.rs
+++ b/crates/language_models/src/provider/vercel.rs
@@ -355,6 +355,7 @@ impl LanguageModel for VercelLanguageModel {
             request,
             self.model.id(),
             self.model.supports_parallel_tool_calls(),
+            self.model.supports_prompt_cache_key(),
             self.max_output_tokens(),
             None,
         );
diff --git a/crates/language_models/src/provider/x_ai.rs b/crates/language_models/src/provider/x_ai.rs
index 5e7190ea96..fed6fe92bf 100644
--- a/crates/language_models/src/provider/x_ai.rs
+++ b/crates/language_models/src/provider/x_ai.rs
@@ -359,6 +359,7 @@ impl LanguageModel for XAiLanguageModel {
             request,
             self.model.id(),
             self.model.supports_parallel_tool_calls(),
+            self.model.supports_prompt_cache_key(),
             self.max_output_tokens(),
             None,
         );
diff --git a/crates/open_ai/src/open_ai.rs b/crates/open_ai/src/open_ai.rs
index 8bbe858995..604e8fe622 100644
--- a/crates/open_ai/src/open_ai.rs
+++ b/crates/open_ai/src/open_ai.rs
@@ -236,6 +236,13 @@ impl Model {
             Self::O1 | Self::O3 | Self::O3Mini | Self::O4Mini | Model::Custom { .. } => false,
         }
     }
+
+    /// Returns whether the given model supports the `prompt_cache_key` parameter.
+    ///
+    /// If the model does not support the parameter, do not pass it up.
+    pub fn supports_prompt_cache_key(&self) -> bool {
+        return true;
+    }
 }
 
 #[derive(Debug, Serialize, Deserialize)]
diff --git a/crates/vercel/src/vercel.rs b/crates/vercel/src/vercel.rs
index 1ae22c5fef..8686fda53f 100644
--- a/crates/vercel/src/vercel.rs
+++ b/crates/vercel/src/vercel.rs
@@ -71,4 +71,8 @@ impl Model {
             Model::Custom { .. } => false,
         }
     }
+
+    pub fn supports_prompt_cache_key(&self) -> bool {
+        false
+    }
 }
diff --git a/crates/x_ai/src/x_ai.rs b/crates/x_ai/src/x_ai.rs
index ac116b2f8f..23cd5b9320 100644
--- a/crates/x_ai/src/x_ai.rs
+++ b/crates/x_ai/src/x_ai.rs
@@ -105,6 +105,10 @@ impl Model {
         }
     }
 
+    pub fn supports_prompt_cache_key(&self) -> bool {
+        false
+    }
+
     pub fn supports_tool(&self) -> bool {
         match self {
             Self::Grok2Vision