diff --git a/codex-rs/codex-api/src/endpoint/models.rs b/codex-rs/codex-api/src/endpoint/models.rs
index b15f07fca2a..7f21c776352 100644
--- a/codex-rs/codex-api/src/endpoint/models.rs
+++ b/codex-rs/codex-api/src/endpoint/models.rs
@@ -5,6 +5,7 @@ use crate::provider::Provider;
 use crate::telemetry::run_with_request_telemetry;
 use codex_client::HttpTransport;
 use codex_client::RequestTelemetry;
+use codex_protocol::openai_models::ModelInfo;
 use codex_protocol::openai_models::ModelsResponse;
 use http::HeaderMap;
 use http::Method;
@@ -41,7 +42,7 @@ impl<T: HttpTransport, A: AuthProvider> ModelsClient<T, A> {
         &self,
         client_version: &str,
         extra_headers: HeaderMap,
-    ) -> Result<ModelsResponse, ApiError> {
+    ) -> Result<(Vec<ModelInfo>, Option<String>), ApiError> {
         let builder = || {
             let mut req = self.provider.build_request(Method::GET, self.path());
             req.headers.extend(extra_headers.clone());
@@ -66,7 +67,7 @@ impl<T: HttpTransport, A: AuthProvider> ModelsClient<T, A> {
             .and_then(|value| value.to_str().ok())
             .map(ToString::to_string);
 
-        let ModelsResponse { models, etag } = serde_json::from_slice::<ModelsResponse>(&resp.body)
+        let ModelsResponse { models } = serde_json::from_slice::<ModelsResponse>(&resp.body)
             .map_err(|e| {
                 ApiError::Stream(format!(
                     "failed to decode models response: {e}; body: {}",
@@ -74,9 +75,7 @@ impl<T: HttpTransport, A: AuthProvider> ModelsClient<T, A> {
                 ))
             })?;
 
-        let etag = header_etag.unwrap_or(etag);
-
-        Ok(ModelsResponse { models, etag })
+        Ok((models, header_etag))
     }
 }
 
@@ -102,16 +101,15 @@ mod tests {
     struct CapturingTransport {
         last_request: Arc<Mutex<Option<Request>>>,
         body: Arc<ModelsResponse>,
+        response_etag: Arc<Option<String>>,
     }
 
     impl Default for CapturingTransport {
         fn default() -> Self {
             Self {
                 last_request: Arc::new(Mutex::new(None)),
-                body: Arc::new(ModelsResponse {
-                    models: Vec::new(),
-                    etag: String::new(),
-                }),
+                body: Arc::new(ModelsResponse { models: Vec::new() }),
+                response_etag: Arc::new(None),
             }
         }
     }
@@ -122,8 +120,8 @@ mod tests {
             *self.last_request.lock().unwrap() = Some(req);
             let body = serde_json::to_vec(&*self.body).unwrap();
             let mut headers = HeaderMap::new();
-            if !self.body.etag.is_empty() {
-                headers.insert(ETAG, self.body.etag.parse().unwrap());
+            if let Some(etag) = self.response_etag.as_ref().as_deref() {
+                headers.insert(ETAG, etag.parse().unwrap());
             }
             Ok(Response {
                 status: StatusCode::OK,
@@ -166,14 +164,12 @@ mod tests {
 
     #[tokio::test]
     async fn appends_client_version_query() {
-        let response = ModelsResponse {
-            models: Vec::new(),
-            etag: String::new(),
-        };
+        let response = ModelsResponse { models: Vec::new() };
 
         let transport = CapturingTransport {
             last_request: Arc::new(Mutex::new(None)),
             body: Arc::new(response),
+            response_etag: Arc::new(None),
         };
 
         let client = ModelsClient::new(
@@ -182,12 +178,12 @@ mod tests {
             DummyAuth,
         );
 
-        let result = client
+        let (models, _etag) = client
             .list_models("0.99.0", HeaderMap::new())
             .await
             .expect("request should succeed");
 
-        assert_eq!(result.models.len(), 0);
+        assert_eq!(models.len(), 0);
 
         let url = transport
             .last_request
@@ -232,12 +228,12 @@ mod tests {
                 }))
                 .unwrap(),
             ],
-            etag: String::new(),
         };
 
         let transport = CapturingTransport {
             last_request: Arc::new(Mutex::new(None)),
             body: Arc::new(response),
+            response_etag: Arc::new(None),
         };
 
         let client = ModelsClient::new(
@@ -246,27 +242,25 @@ mod tests {
             DummyAuth,
         );
 
-        let result = client
+        let (models, _etag) = client
             .list_models("0.99.0", HeaderMap::new())
             .await
             .expect("request should succeed");
 
-        assert_eq!(result.models.len(), 1);
-        assert_eq!(result.models[0].slug, "gpt-test");
-        assert_eq!(result.models[0].supported_in_api, true);
-        assert_eq!(result.models[0].priority, 1);
+        assert_eq!(models.len(), 1);
+        assert_eq!(models[0].slug, "gpt-test");
+        assert_eq!(models[0].supported_in_api, true);
+        assert_eq!(models[0].priority, 1);
     }
 
     #[tokio::test]
     async fn list_models_includes_etag() {
-        let response = ModelsResponse {
-            models: Vec::new(),
-            etag: "\"abc\"".to_string(),
-        };
+        let response = ModelsResponse { models: Vec::new() };
 
         let transport = CapturingTransport {
             last_request: Arc::new(Mutex::new(None)),
             body: Arc::new(response),
+            response_etag: Arc::new(Some("\"abc\"".to_string())),
         };
 
         let client = ModelsClient::new(
@@ -275,12 +269,12 @@ mod tests {
             DummyAuth,
         );
 
-        let result = client
+        let (models, etag) = client
             .list_models("0.1.0", HeaderMap::new())
             .await
             .expect("request should succeed");
 
-        assert_eq!(result.models.len(), 0);
-        assert_eq!(result.etag, "\"abc\"");
+        assert_eq!(models.len(), 0);
+        assert_eq!(etag.as_deref(), Some("\"abc\""));
     }
 }
diff --git a/codex-rs/codex-api/tests/models_integration.rs b/codex-rs/codex-api/tests/models_integration.rs
index 93baffd3560..0b3e95ee303 100644
--- a/codex-rs/codex-api/tests/models_integration.rs
+++ b/codex-rs/codex-api/tests/models_integration.rs
@@ -90,7 +90,6 @@ async fn models_client_hits_models_endpoint() {
             reasoning_summary_format: ReasoningSummaryFormat::None,
             experimental_supported_tools: Vec::new(),
         }],
-        etag: String::new(),
     };
 
     Mock::given(method("GET"))
@@ -106,13 +105,13 @@ async fn models_client_hits_models_endpoint() {
     let transport = ReqwestTransport::new(reqwest::Client::new());
     let client = ModelsClient::new(transport, provider(&base_url), DummyAuth);
 
-    let result = client
+    let (models, _etag) = client
         .list_models("0.1.0", HeaderMap::new())
         .await
         .expect("models request should succeed");
 
-    assert_eq!(result.models.len(), 1);
-    assert_eq!(result.models[0].slug, "gpt-test");
+    assert_eq!(models.len(), 1);
+    assert_eq!(models[0].slug, "gpt-test");
 
     let received = server
         .received_requests()
diff --git a/codex-rs/core/src/api_bridge.rs b/codex-rs/core/src/api_bridge.rs
index 79fd67d6501..c7da55da853 100644
--- a/codex-rs/core/src/api_bridge.rs
+++ b/codex-rs/core/src/api_bridge.rs
@@ -67,6 +67,11 @@ pub(crate) fn map_api_error(err: ApiError) -> CodexErr {
                         status,
                         request_id: extract_request_id(headers.as_ref()),
                     })
+                } else if status == http::StatusCode::PRECONDITION_FAILED
+                    && body_text
+                        .contains("Models catalog has changed. Please refresh your models list.")
+                {
+                    CodexErr::OutdatedModels
                 } else {
                     CodexErr::UnexpectedStatus(UnexpectedResponseError {
                         status,
diff --git a/codex-rs/core/src/client.rs b/codex-rs/core/src/client.rs
index aaf3b0ea353..3426da78c17 100644
--- a/codex-rs/core/src/client.rs
+++ b/codex-rs/core/src/client.rs
@@ -33,6 +33,7 @@ use http::StatusCode as HttpStatusCode;
 use reqwest::StatusCode;
 use serde_json::Value;
 use std::time::Duration;
+use tokio::sync::RwLock;
 use tokio::sync::mpsc;
 use tracing::warn;
 
@@ -53,11 +54,12 @@ use crate::openai_models::model_family::ModelFamily;
 use crate::tools::spec::create_tools_json_for_chat_completions_api;
 use crate::tools::spec::create_tools_json_for_responses_api;
 
-#[derive(Debug, Clone)]
+#[derive(Debug)]
 pub struct ModelClient {
     config: Arc<Config>,
     auth_manager: Option<Arc<AuthManager>>,
-    model_family: ModelFamily,
+    model_family: RwLock<ModelFamily>,
+    models_etag: RwLock<Option<String>>,
     otel_manager: OtelManager,
     provider: ModelProviderInfo,
     conversation_id: ConversationId,
@@ -72,6 +74,7 @@ impl ModelClient {
         config: Arc<Config>,
         auth_manager: Option<Arc<AuthManager>>,
         model_family: ModelFamily,
+        models_etag: Option<String>,
         otel_manager: OtelManager,
         provider: ModelProviderInfo,
         effort: Option<ReasoningEffortConfig>,
@@ -82,7 +85,8 @@ impl ModelClient {
         Self {
             config,
             auth_manager,
-            model_family,
+            model_family: RwLock::new(model_family),
+            models_etag: RwLock::new(models_etag),
             otel_manager,
             provider,
             conversation_id,
@@ -92,8 +96,8 @@ impl ModelClient {
         }
     }
 
-    pub fn get_model_context_window(&self) -> Option<i64> {
-        let model_family = self.get_model_family();
+    pub async fn get_model_context_window(&self) -> Option<i64> {
+        let model_family = self.get_model_family().await;
         let effective_context_window_percent = model_family.effective_context_window_percent;
         model_family
             .context_window
@@ -146,7 +150,7 @@ impl ModelClient {
         }
 
         let auth_manager = self.auth_manager.clone();
-        let model_family = self.get_model_family();
+        let model_family = self.get_model_family().await;
         let instructions = prompt.get_full_instructions(&model_family).into_owned();
         let tools_json = create_tools_json_for_chat_completions_api(&prompt.tools)?;
         let api_prompt = build_api_prompt(prompt, instructions, tools_json);
@@ -167,7 +171,7 @@ impl ModelClient {
 
             let stream_result = client
                 .stream_prompt(
-                    &self.get_model(),
+                    &self.get_model().await,
                     &api_prompt,
                     Some(conversation_id.clone()),
                     Some(session_source.clone()),
@@ -200,7 +204,7 @@ impl ModelClient {
         }
 
         let auth_manager = self.auth_manager.clone();
-        let model_family = self.get_model_family();
+        let model_family = self.get_model_family().await;
         let instructions = prompt.get_full_instructions(&model_family).into_owned();
         let tools_json: Vec<Value> = create_tools_json_for_responses_api(&prompt.tools)?;
 
@@ -262,11 +266,14 @@ impl ModelClient {
                 store_override: None,
                 conversation_id: Some(conversation_id.clone()),
                 session_source: Some(session_source.clone()),
-                extra_headers: beta_feature_headers(&self.config),
+                extra_headers: beta_feature_headers(
+                    &self.config,
+                    self.get_models_etag().await.clone(),
+                ),
             };
 
             let stream_result = client
-                .stream_prompt(&self.get_model(), &api_prompt, options)
+                .stream_prompt(&self.get_model().await, &api_prompt, options)
                 .await;
 
             match stream_result {
@@ -297,13 +304,25 @@ impl ModelClient {
     }
 
     /// Returns the currently configured model slug.
-    pub fn get_model(&self) -> String {
-        self.get_model_family().get_model_slug().to_string()
+    pub async fn get_model(&self) -> String {
+        self.get_model_family().await.get_model_slug().to_string()
     }
 
     /// Returns the currently configured model family.
-    pub fn get_model_family(&self) -> ModelFamily {
-        self.model_family.clone()
+    pub async fn get_model_family(&self) -> ModelFamily {
+        self.model_family.read().await.clone()
+    }
+
+    pub async fn get_models_etag(&self) -> Option<String> {
+        self.models_etag.read().await.clone()
+    }
+
+    pub async fn update_models_etag(&self, etag: Option<String>) {
+        *self.models_etag.write().await = etag;
+    }
+
+    pub async fn update_model_family(&self, model_family: ModelFamily) {
+        *self.model_family.write().await = model_family;
     }
 
     /// Returns the current reasoning effort setting.
@@ -340,10 +359,10 @@ impl ModelClient {
             .with_telemetry(Some(request_telemetry));
 
         let instructions = prompt
-            .get_full_instructions(&self.get_model_family())
+            .get_full_instructions(&self.get_model_family().await)
             .into_owned();
         let payload = ApiCompactionInput {
-            model: &self.get_model(),
+            model: &self.get_model().await,
             input: &prompt.input,
             instructions: &instructions,
         };
@@ -398,7 +417,7 @@ fn build_api_prompt(prompt: &Prompt, instructions: String, tools_json: Vec<Value
     }
 }
 
-fn beta_feature_headers(config: &Config) -> ApiHeaderMap {
+fn beta_feature_headers(config: &Config, models_etag: Option<String>) -> ApiHeaderMap {
     let enabled = FEATURES
         .iter()
         .filter_map(|spec| {
@@ -416,6 +435,11 @@ fn beta_feature_headers(config: &Config) -> ApiHeaderMap {
     {
         headers.insert("x-codex-beta-features", header_value);
     }
+    if let Some(etag) = models_etag
+        && let Ok(header_value) = HeaderValue::from_str(&etag)
+    {
+        headers.insert("X-If-Models-Match", header_value);
+    }
     headers
 }
 
diff --git a/codex-rs/core/src/client_common.rs b/codex-rs/core/src/client_common.rs
index 4a3bc8de235..db1a7333d18 100644
--- a/codex-rs/core/src/client_common.rs
+++ b/codex-rs/core/src/client_common.rs
@@ -1,6 +1,10 @@
 use crate::client_common::tools::ToolSpec;
+use crate::codex::Session;
+use crate::codex::TurnContext;
 use crate::error::Result;
+use crate::features::Feature;
 use crate::openai_models::model_family::ModelFamily;
+use crate::tools::ToolRouter;
 pub use codex_api::common::ResponseEvent;
 use codex_apply_patch::APPLY_PATCH_TOOL_INSTRUCTIONS;
 use codex_protocol::models::ResponseItem;
@@ -44,6 +48,28 @@ pub struct Prompt {
 }
 
 impl Prompt {
+    pub(crate) async fn new(
+        sess: &Session,
+        turn_context: &TurnContext,
+        router: &ToolRouter,
+        input: &[ResponseItem],
+    ) -> Prompt {
+        let model_supports_parallel = turn_context
+            .client
+            .get_model_family()
+            .await
+            .supports_parallel_tool_calls;
+
+        Prompt {
+            input: input.to_vec(),
+            tools: router.specs(),
+            parallel_tool_calls: model_supports_parallel
+                && sess.enabled(Feature::ParallelToolCalls),
+            base_instructions_override: turn_context.base_instructions.clone(),
+            output_schema: turn_context.final_output_json_schema.clone(),
+        }
+    }
+
     pub(crate) fn get_full_instructions<'a>(&'a self, model: &'a ModelFamily) -> Cow<'a, str> {
         let base = self
             .base_instructions_override
diff --git a/codex-rs/core/src/codex.rs b/codex-rs/core/src/codex.rs
index f0d2056587c..0c6b5739894 100644
--- a/codex-rs/core/src/codex.rs
+++ b/codex-rs/core/src/codex.rs
@@ -249,7 +249,7 @@ impl Codex {
 
         let config = Arc::new(config);
         if config.features.enabled(Feature::RemoteModels)
-            && let Err(err) = models_manager.refresh_available_models(&config).await
+            && let Err(err) = models_manager.try_refresh_available_models(&config).await
         {
             error!("failed to refresh available models: {err:?}");
         }
@@ -492,6 +492,7 @@ impl Session {
         session_configuration: &SessionConfiguration,
         per_turn_config: Config,
         model_family: ModelFamily,
+        models_etag: Option<String>,
         conversation_id: ConversationId,
         sub_id: String,
     ) -> TurnContext {
@@ -505,6 +506,7 @@ impl Session {
             per_turn_config.clone(),
             auth_manager,
             model_family.clone(),
+            models_etag,
             otel_manager,
             provider,
             session_configuration.model_reasoning_effort,
@@ -788,7 +790,7 @@ impl Session {
                         }
                     })
                 {
-                    let curr = turn_context.client.get_model();
+                    let curr = turn_context.client.get_model().await;
                     if prev != curr {
                         warn!(
                             "resuming session with different model: previous={prev}, current={curr}"
@@ -919,6 +921,7 @@ impl Session {
             .models_manager
             .construct_model_family(session_configuration.model.as_str(), &per_turn_config)
             .await;
+        let models_etag = self.services.models_manager.get_models_etag().await;
         let mut turn_context: TurnContext = Self::make_turn_context(
             Some(Arc::clone(&self.services.auth_manager)),
             &self.services.otel_manager,
@@ -926,6 +929,7 @@ impl Session {
             &session_configuration,
             per_turn_config,
             model_family,
+            models_etag,
             self.conversation_id,
             sub_id,
         );
@@ -1334,7 +1338,7 @@ impl Session {
             if let Some(token_usage) = token_usage {
                 state.update_token_info_from_usage(
                     token_usage,
-                    turn_context.client.get_model_context_window(),
+                    turn_context.client.get_model_context_window().await,
                 );
             }
         }
@@ -1346,6 +1350,7 @@ impl Session {
             .clone_history()
             .await
             .estimate_token_count(turn_context)
+            .await
         else {
             return;
         };
@@ -1366,7 +1371,7 @@ impl Session {
             };
 
             if info.model_context_window.is_none() {
-                info.model_context_window = turn_context.client.get_model_context_window();
+                info.model_context_window = turn_context.client.get_model_context_window().await;
             }
 
             state.set_token_info(Some(info));
@@ -1396,7 +1401,7 @@ impl Session {
     }
 
     pub(crate) async fn set_total_tokens_full(&self, turn_context: &TurnContext) {
-        let context_window = turn_context.client.get_model_context_window();
+        let context_window = turn_context.client.get_model_context_window().await;
         if let Some(context_window) = context_window {
             {
                 let mut state = self.state.lock().await;
@@ -2105,6 +2110,7 @@ async fn spawn_review_thread(
         .models_manager
         .construct_model_family(&model, &config)
         .await;
+    let models_etag = sess.services.models_manager.get_models_etag().await;
     // For reviews, disable web_search and view_image regardless of global settings.
     let mut review_features = sess.features.clone();
     review_features
@@ -2137,6 +2143,7 @@ async fn spawn_review_thread(
         per_turn_config.clone(),
         auth_manager,
         model_family.clone(),
+        models_etag,
         otel_manager,
         provider,
         per_turn_config.model_reasoning_effort,
@@ -2231,6 +2238,7 @@ pub(crate) async fn run_task(
     let auto_compact_limit = turn_context
         .client
         .get_model_family()
+        .await
         .auto_compact_token_limit()
         .unwrap_or(i64::MAX);
     let total_usage_tokens = sess.get_total_token_usage().await;
@@ -2238,7 +2246,7 @@ pub(crate) async fn run_task(
         run_auto_compact(&sess, &turn_context).await;
     }
     let event = EventMsg::TaskStarted(TaskStartedEvent {
-        model_context_window: turn_context.client.get_model_context_window(),
+        model_context_window: turn_context.client.get_model_context_window().await,
     });
     sess.send_event(&turn_context, event).await;
 
@@ -2303,7 +2311,7 @@ pub(crate) async fn run_task(
             .collect::<Vec<String>>();
         match run_turn(
             Arc::clone(&sess),
-            Arc::clone(&turn_context),
+            &turn_context,
             Arc::clone(&turn_diff_tracker),
             turn_input,
             cancellation_token.child_token(),
@@ -2362,6 +2370,36 @@ pub(crate) async fn run_task(
     last_agent_message
 }
 
+pub(crate) async fn refresh_models_and_reset_turn_context(
+    sess: &Arc<Session>,
+    turn_context: &Arc<TurnContext>,
+) {
+    let config = {
+        let state = sess.state.lock().await;
+        state
+            .session_configuration
+            .original_config_do_not_use
+            .clone()
+    };
+    if let Err(err) = sess
+        .services
+        .models_manager
+        .refresh_available_models(&config)
+        .await
+    {
+        error!("failed to refresh models after outdated models error: {err}");
+    }
+    let model = turn_context.client.get_model().await;
+    let model_family = sess
+        .services
+        .models_manager
+        .construct_model_family(&model, &config)
+        .await;
+    let models_etag = sess.services.models_manager.get_models_etag().await;
+    turn_context.client.update_model_family(model_family).await;
+    turn_context.client.update_models_etag(models_etag).await;
+}
+
 async fn run_auto_compact(sess: &Arc<Session>, turn_context: &Arc<TurnContext>) {
     if should_use_remote_compact_task(sess.as_ref(), &turn_context.client.get_provider()) {
         run_inline_remote_auto_compact_task(Arc::clone(sess), Arc::clone(turn_context)).await;
@@ -2374,17 +2412,19 @@ async fn run_auto_compact(sess: &Arc<Session>, turn_context: &Arc<TurnContext>)
     skip_all,
     fields(
         turn_id = %turn_context.sub_id,
-        model = %turn_context.client.get_model(),
+        model = tracing::field::Empty,
         cwd = %turn_context.cwd.display()
     )
 )]
 async fn run_turn(
     sess: Arc<Session>,
-    turn_context: Arc<TurnContext>,
+    turn_context: &Arc<TurnContext>,
     turn_diff_tracker: SharedTurnDiffTracker,
     input: Vec<ResponseItem>,
     cancellation_token: CancellationToken,
 ) -> CodexResult<TurnRunResult> {
+    let model = turn_context.client.get_model().await;
+    tracing::Span::current().record("model", field::display(&model));
     let mcp_tools = sess
         .services
         .mcp_connection_manager
@@ -2393,37 +2433,32 @@ async fn run_turn(
         .list_all_tools()
         .or_cancel(&cancellation_token)
         .await?;
-    let router = Arc::new(ToolRouter::from_config(
-        &turn_context.tools_config,
-        Some(
-            mcp_tools
-                .into_iter()
-                .map(|(name, tool)| (name, tool.tool))
-                .collect(),
-        ),
-    ));
-
-    let model_supports_parallel = turn_context
-        .client
-        .get_model_family()
-        .supports_parallel_tool_calls;
-
-    let prompt = Prompt {
-        input,
-        tools: router.specs(),
-        parallel_tool_calls: model_supports_parallel && sess.enabled(Feature::ParallelToolCalls),
-        base_instructions_override: turn_context.base_instructions.clone(),
-        output_schema: turn_context.final_output_json_schema.clone(),
-    };
 
     let mut retries = 0;
     loop {
+        let router = Arc::new(ToolRouter::from_config(
+            &turn_context.tools_config,
+            Some(
+                mcp_tools
+                    .clone()
+                    .into_iter()
+                    .map(|(name, tool)| (name, tool.tool))
+                    .collect(),
+            ),
+        ));
+        let prompt = Prompt::new(
+            sess.as_ref(),
+            turn_context.as_ref(),
+            router.as_ref(),
+            &input,
+        );
+
         match try_run_turn(
             Arc::clone(&router),
             Arc::clone(&sess),
-            Arc::clone(&turn_context),
+            Arc::clone(turn_context),
             Arc::clone(&turn_diff_tracker),
-            &prompt,
+            &prompt.await,
             cancellation_token.child_token(),
         )
         .await
@@ -2437,13 +2472,13 @@ async fn run_turn(
             Err(CodexErr::EnvVar(var)) => return Err(CodexErr::EnvVar(var)),
             Err(e @ CodexErr::Fatal(_)) => return Err(e),
             Err(e @ CodexErr::ContextWindowExceeded) => {
-                sess.set_total_tokens_full(&turn_context).await;
+                sess.set_total_tokens_full(turn_context).await;
                 return Err(e);
             }
             Err(CodexErr::UsageLimitReached(e)) => {
                 let rate_limits = e.rate_limits.clone();
                 if let Some(rate_limits) = rate_limits {
-                    sess.update_rate_limits(&turn_context, rate_limits).await;
+                    sess.update_rate_limits(turn_context, rate_limits).await;
                 }
                 return Err(CodexErr::UsageLimitReached(e));
             }
@@ -2457,6 +2492,11 @@ async fn run_turn(
                 let max_retries = turn_context.client.get_provider().stream_max_retries();
                 if retries < max_retries {
                     retries += 1;
+                    // Refresh models if we got an outdated models error
+                    if matches!(e, CodexErr::OutdatedModels) {
+                        refresh_models_and_reset_turn_context(&sess, turn_context).await;
+                        continue;
+                    }
                     let delay = match e {
                         CodexErr::Stream(_, Some(delay)) => delay,
                         _ => backoff(retries),
@@ -2469,7 +2509,7 @@ async fn run_turn(
                     // user understands what is happening instead of staring
                     // at a seemingly frozen screen.
                     sess.notify_stream_error(
-                        &turn_context,
+                        turn_context,
                         format!("Reconnecting... {retries}/{max_retries}"),
                         e,
                     )
@@ -2514,7 +2554,7 @@ async fn drain_in_flight(
     skip_all,
     fields(
         turn_id = %turn_context.sub_id,
-        model = %turn_context.client.get_model()
+        model = tracing::field::Empty,
     )
 )]
 async fn try_run_turn(
@@ -2525,11 +2565,13 @@ async fn try_run_turn(
     prompt: &Prompt,
     cancellation_token: CancellationToken,
 ) -> CodexResult<TurnRunResult> {
+    let model = turn_context.client.get_model().await;
+    tracing::Span::current().record("model", field::display(&model));
     let rollout_item = RolloutItem::TurnContext(TurnContextItem {
         cwd: turn_context.cwd.clone(),
         approval_policy: turn_context.approval_policy,
         sandbox_policy: turn_context.sandbox_policy.clone(),
-        model: turn_context.client.get_model(),
+        model,
         effort: turn_context.client.get_reasoning_effort(),
         summary: turn_context.client.get_reasoning_summary(),
     });
@@ -2537,7 +2579,6 @@ async fn try_run_turn(
     sess.persist_rollout_items(&[rollout_item]).await;
     let mut stream = turn_context
         .client
-        .clone()
         .stream(prompt)
         .instrument(trace_span!("stream_request"))
         .or_cancel(&cancellation_token)
@@ -3163,6 +3204,7 @@ mod tests {
             &session_configuration,
             per_turn_config,
             model_family,
+            None,
             conversation_id,
             "turn_id".to_string(),
         );
@@ -3249,6 +3291,7 @@ mod tests {
             &session_configuration,
             per_turn_config,
             model_family,
+            None,
             conversation_id,
             "turn_id".to_string(),
         ));
diff --git a/codex-rs/core/src/compact.rs b/codex-rs/core/src/compact.rs
index 1a90b7b223f..a864774d69c 100644
--- a/codex-rs/core/src/compact.rs
+++ b/codex-rs/core/src/compact.rs
@@ -6,6 +6,7 @@ use crate::client_common::ResponseEvent;
 use crate::codex::Session;
 use crate::codex::TurnContext;
 use crate::codex::get_last_assistant_message_from_turn;
+use crate::codex::refresh_models_and_reset_turn_context;
 use crate::error::CodexErr;
 use crate::error::Result as CodexResult;
 use crate::features::Feature;
@@ -55,7 +56,7 @@ pub(crate) async fn run_compact_task(
     input: Vec<UserInput>,
 ) {
     let start_event = EventMsg::TaskStarted(TaskStartedEvent {
-        model_context_window: turn_context.client.get_model_context_window(),
+        model_context_window: turn_context.client.get_model_context_window().await,
     });
     sess.send_event(&turn_context, start_event).await;
     run_compact_task_inner(sess.clone(), turn_context, input).await;
@@ -83,7 +84,7 @@ async fn run_compact_task_inner(
         cwd: turn_context.cwd.clone(),
         approval_policy: turn_context.approval_policy,
         sandbox_policy: turn_context.sandbox_policy.clone(),
-        model: turn_context.client.get_model(),
+        model: turn_context.client.get_model().await,
         effort: turn_context.client.get_reasoning_effort(),
         summary: turn_context.client.get_reasoning_summary(),
     });
@@ -132,6 +133,10 @@ async fn run_compact_task_inner(
             Err(e) => {
                 if retries < max_retries {
                     retries += 1;
+                    if matches!(e, CodexErr::OutdatedModels) {
+                        refresh_models_and_reset_turn_context(&sess, &turn_context).await;
+                        continue;
+                    }
                     let delay = backoff(retries);
                     sess.notify_stream_error(
                         turn_context.as_ref(),
@@ -290,7 +295,7 @@ async fn drain_to_completed(
     turn_context: &TurnContext,
     prompt: &Prompt,
 ) -> CodexResult<()> {
-    let mut stream = turn_context.client.clone().stream(prompt).await?;
+    let mut stream = turn_context.client.stream(prompt).await?;
     loop {
         let maybe_event = stream.next().await;
         let Some(event) = maybe_event else {
diff --git a/codex-rs/core/src/compact_remote.rs b/codex-rs/core/src/compact_remote.rs
index b855f28d39d..3419b2e51fd 100644
--- a/codex-rs/core/src/compact_remote.rs
+++ b/codex-rs/core/src/compact_remote.rs
@@ -20,7 +20,7 @@ pub(crate) async fn run_inline_remote_auto_compact_task(
 
 pub(crate) async fn run_remote_compact_task(sess: Arc<Session>, turn_context: Arc<TurnContext>) {
     let start_event = EventMsg::TaskStarted(TaskStartedEvent {
-        model_context_window: turn_context.client.get_model_context_window(),
+        model_context_window: turn_context.client.get_model_context_window().await,
     });
     sess.send_event(&turn_context, start_event).await;
 
diff --git a/codex-rs/core/src/context_manager/history.rs b/codex-rs/core/src/context_manager/history.rs
index c18ad7df8ec..841be4bae58 100644
--- a/codex-rs/core/src/context_manager/history.rs
+++ b/codex-rs/core/src/context_manager/history.rs
@@ -79,8 +79,8 @@ impl ContextManager {
 
     // Estimate token usage using byte-based heuristics from the truncation helpers.
     // This is a coarse lower bound, not a tokenizer-accurate count.
-    pub(crate) fn estimate_token_count(&self, turn_context: &TurnContext) -> Option<i64> {
-        let model_family = turn_context.client.get_model_family();
+    pub(crate) async fn estimate_token_count(&self, turn_context: &TurnContext) -> Option<i64> {
+        let model_family = turn_context.client.get_model_family().await;
         let base_tokens =
             i64::try_from(approx_token_count(model_family.base_instructions.as_str()))
                 .unwrap_or(i64::MAX);
diff --git a/codex-rs/core/src/error.rs b/codex-rs/core/src/error.rs
index e8fa91d26e8..e85eaf0dbfe 100644
--- a/codex-rs/core/src/error.rs
+++ b/codex-rs/core/src/error.rs
@@ -90,6 +90,10 @@ pub enum CodexErr {
     #[error("spawn failed: child stdout/stderr not captured")]
     Spawn,
 
+    /// Returned when the models list is outdated and needs to be refreshed.
+    #[error("remote models list is outdated")]
+    OutdatedModels,
+
     /// Returned by run_command_stream when the user pressed Ctrl‑C (SIGINT). Session uses this to
     /// surface a polite FunctionCallOutput back to the model instead of crashing the CLI.
     #[error("interrupted (Ctrl-C). Something went wrong? Hit `/feedback` to report the issue.")]
diff --git a/codex-rs/core/src/openai_models/models_manager.rs b/codex-rs/core/src/openai_models/models_manager.rs
index 7f54c4f8525..999ac20164d 100644
--- a/codex-rs/core/src/openai_models/models_manager.rs
+++ b/codex-rs/core/src/openai_models/models_manager.rs
@@ -77,7 +77,7 @@ impl ModelsManager {
     }
 
     /// Fetch the latest remote models, using the on-disk cache when still fresh.
-    pub async fn refresh_available_models(&self, config: &Config) -> CoreResult<()> {
+    pub async fn try_refresh_available_models(&self, config: &Config) -> CoreResult<()> {
         if !config.features.enabled(Feature::RemoteModels)
             || self.auth_manager.get_auth_mode() == Some(AuthMode::ApiKey)
         {
@@ -86,7 +86,15 @@ impl ModelsManager {
         if self.try_load_cache().await {
             return Ok(());
         }
+        self.refresh_available_models(config).await
+    }
 
+    pub async fn refresh_available_models(&self, config: &Config) -> CoreResult<()> {
+        if !config.features.enabled(Feature::RemoteModels)
+            || self.auth_manager.get_auth_mode() == Some(AuthMode::ApiKey)
+        {
+            return Ok(());
+        }
         let auth = self.auth_manager.auth();
         let api_provider = self.provider.to_api_provider(Some(AuthMode::ChatGPT))?;
         let api_auth = auth_provider_from_auth(auth.clone(), &self.provider).await?;
@@ -94,12 +102,12 @@ impl ModelsManager {
         let client = ModelsClient::new(transport, api_provider, api_auth);
 
         let client_version = format_client_version_to_whole();
-        let ModelsResponse { models, etag } = client
+        let (models, etag) = client
             .list_models(&client_version, HeaderMap::new())
             .await
             .map_err(map_api_error)?;
 
-        let etag = (!etag.is_empty()).then_some(etag);
+        let etag = etag.filter(|value| !value.is_empty());
 
         self.apply_remote_models(models.clone()).await;
         *self.etag.write().await = etag.clone();
@@ -108,7 +116,7 @@ impl ModelsManager {
     }
 
     pub async fn list_models(&self, config: &Config) -> Vec<ModelPreset> {
-        if let Err(err) = self.refresh_available_models(config).await {
+        if let Err(err) = self.try_refresh_available_models(config).await {
             error!("failed to refresh available models: {err}");
         }
         let remote_models = self.remote_models(config).await;
@@ -131,11 +139,15 @@ impl ModelsManager {
             .with_config_overrides(config)
     }
 
+    pub async fn get_models_etag(&self) -> Option<String> {
+        self.etag.read().await.clone()
+    }
+
     pub async fn get_model(&self, model: &Option<String>, config: &Config) -> String {
         if let Some(model) = model.as_ref() {
             return model.to_string();
         }
-        if let Err(err) = self.refresh_available_models(config).await {
+        if let Err(err) = self.try_refresh_available_models(config).await {
             error!("failed to refresh available models: {err}");
         }
         // if codex-auto-balanced exists & signed in with chatgpt mode, return it, otherwise return the default model
@@ -389,7 +401,6 @@ mod tests {
             &server,
             ModelsResponse {
                 models: remote_models.clone(),
-                etag: String::new(),
             },
         )
         .await;
@@ -407,7 +418,7 @@ mod tests {
         let manager = ModelsManager::with_provider(auth_manager, provider);
 
         manager
-            .refresh_available_models(&config)
+            .try_refresh_available_models(&config)
             .await
             .expect("refresh succeeds");
         let cached_remote = manager.remote_models(&config).await;
@@ -446,7 +457,6 @@ mod tests {
             &server,
             ModelsResponse {
                 models: remote_models.clone(),
-                etag: String::new(),
             },
         )
         .await;
@@ -467,7 +477,7 @@ mod tests {
         let manager = ModelsManager::with_provider(auth_manager, provider);
 
         manager
-            .refresh_available_models(&config)
+            .try_refresh_available_models(&config)
             .await
             .expect("first refresh succeeds");
         assert_eq!(
@@ -478,7 +488,7 @@ mod tests {
 
         // Second call should read from cache and avoid the network.
         manager
-            .refresh_available_models(&config)
+            .try_refresh_available_models(&config)
             .await
             .expect("cached refresh succeeds");
         assert_eq!(
@@ -501,7 +511,6 @@ mod tests {
             &server,
             ModelsResponse {
                 models: initial_models.clone(),
-                etag: String::new(),
             },
         )
         .await;
@@ -522,7 +531,7 @@ mod tests {
         let manager = ModelsManager::with_provider(auth_manager, provider);
 
         manager
-            .refresh_available_models(&config)
+            .try_refresh_available_models(&config)
             .await
             .expect("initial refresh succeeds");
 
@@ -542,13 +551,12 @@ mod tests {
             &server,
             ModelsResponse {
                 models: updated_models.clone(),
-                etag: String::new(),
             },
         )
         .await;
 
         manager
-            .refresh_available_models(&config)
+            .try_refresh_available_models(&config)
             .await
             .expect("second refresh succeeds");
         assert_eq!(
@@ -576,7 +584,6 @@ mod tests {
             &server,
             ModelsResponse {
                 models: initial_models,
-                etag: String::new(),
             },
         )
         .await;
@@ -595,7 +602,7 @@ mod tests {
         manager.cache_ttl = Duration::ZERO;
 
         manager
-            .refresh_available_models(&config)
+            .try_refresh_available_models(&config)
             .await
             .expect("initial refresh succeeds");
 
@@ -605,13 +612,12 @@ mod tests {
             &server,
             ModelsResponse {
                 models: refreshed_models,
-                etag: String::new(),
             },
         )
         .await;
 
         manager
-            .refresh_available_models(&config)
+            .try_refresh_available_models(&config)
             .await
             .expect("second refresh succeeds");
 
diff --git a/codex-rs/core/src/tasks/user_shell.rs b/codex-rs/core/src/tasks/user_shell.rs
index aec09514ca3..b053020bdb9 100644
--- a/codex-rs/core/src/tasks/user_shell.rs
+++ b/codex-rs/core/src/tasks/user_shell.rs
@@ -59,7 +59,7 @@ impl SessionTask for UserShellCommandTask {
         cancellation_token: CancellationToken,
     ) -> Option<String> {
         let event = EventMsg::TaskStarted(TaskStartedEvent {
-            model_context_window: turn_context.client.get_model_context_window(),
+            model_context_window: turn_context.client.get_model_context_window().await,
         });
         let session = session.clone_session();
         session.send_event(turn_context.as_ref(), event).await;
diff --git a/codex-rs/core/tests/chat_completions_payload.rs b/codex-rs/core/tests/chat_completions_payload.rs
index 5867935470e..b0086ef5186 100644
--- a/codex-rs/core/tests/chat_completions_payload.rs
+++ b/codex-rs/core/tests/chat_completions_payload.rs
@@ -92,6 +92,7 @@ async fn run_request(input: Vec<ResponseItem>) -> Value {
         Arc::clone(&config),
         None,
         model_family,
+        None,
         otel_manager,
         provider,
         effort,
diff --git a/codex-rs/core/tests/chat_completions_sse.rs b/codex-rs/core/tests/chat_completions_sse.rs
index f58b039220e..74f7730504c 100644
--- a/codex-rs/core/tests/chat_completions_sse.rs
+++ b/codex-rs/core/tests/chat_completions_sse.rs
@@ -93,6 +93,7 @@ async fn run_stream_with_bytes(sse_body: &[u8]) -> Vec<ResponseEvent> {
         Arc::clone(&config),
         None,
         model_family,
+        None,
         otel_manager,
         provider,
         effort,
diff --git a/codex-rs/core/tests/common/responses.rs b/codex-rs/core/tests/common/responses.rs
index b98b29625eb..76133a9fa39 100644
--- a/codex-rs/core/tests/common/responses.rs
+++ b/codex-rs/core/tests/common/responses.rs
@@ -670,6 +670,24 @@ pub async fn mount_models_once(server: &MockServer, body: ModelsResponse) -> Mod
     models_mock
 }
 
+pub async fn mount_models_once_with_etag(
+    server: &MockServer,
+    body: ModelsResponse,
+    etag: &str,
+) -> ModelsMock {
+    let (mock, models_mock) = models_mock();
+    mock.respond_with(
+        ResponseTemplate::new(200)
+            .insert_header("content-type", "application/json")
+            .insert_header("etag", etag)
+            .set_body_json(body.clone()),
+    )
+    .up_to_n_times(1)
+    .mount(server)
+    .await;
+    models_mock
+}
+
 pub async fn start_mock_server() -> MockServer {
     let server = MockServer::builder()
         .body_print_limit(BodyPrintLimit::Limited(80_000))
@@ -677,14 +695,7 @@ pub async fn start_mock_server() -> MockServer {
         .await;
 
     // Provide a default `/models` response so tests remain hermetic when the client queries it.
-    let _ = mount_models_once(
-        &server,
-        ModelsResponse {
-            models: Vec::new(),
-            etag: String::new(),
-        },
-    )
-    .await;
+    let _ = mount_models_once(&server, ModelsResponse { models: Vec::new() }).await;
 
     server
 }
diff --git a/codex-rs/core/tests/responses_headers.rs b/codex-rs/core/tests/responses_headers.rs
index 5c32685cc92..f005ec5be9e 100644
--- a/codex-rs/core/tests/responses_headers.rs
+++ b/codex-rs/core/tests/responses_headers.rs
@@ -86,6 +86,7 @@ async fn responses_stream_includes_subagent_header_on_review() {
         Arc::clone(&config),
         None,
         model_family,
+        None,
         otel_manager,
         provider,
         effort,
@@ -181,6 +182,7 @@ async fn responses_stream_includes_subagent_header_on_other() {
         Arc::clone(&config),
         None,
         model_family,
+        None,
         otel_manager,
         provider,
         effort,
@@ -275,6 +277,7 @@ async fn responses_respects_model_family_overrides_from_config() {
         Arc::clone(&config),
         None,
         model_family,
+        None,
         otel_manager,
         provider,
         effort,
diff --git a/codex-rs/core/tests/suite/client.rs b/codex-rs/core/tests/suite/client.rs
index bda232433da..65b1a7f4c8e 100644
--- a/codex-rs/core/tests/suite/client.rs
+++ b/codex-rs/core/tests/suite/client.rs
@@ -1146,6 +1146,7 @@ async fn azure_responses_request_includes_store_and_reasoning_ids() {
         Arc::clone(&config),
         None,
         model_family,
+        None,
         otel_manager,
         provider,
         effort,
diff --git a/codex-rs/core/tests/suite/remote_models.rs b/codex-rs/core/tests/suite/remote_models.rs
index 3c4d389ec05..d54405fe6dc 100644
--- a/codex-rs/core/tests/suite/remote_models.rs
+++ b/codex-rs/core/tests/suite/remote_models.rs
@@ -33,8 +33,12 @@ use core_test_support::responses::ev_assistant_message;
 use core_test_support::responses::ev_completed;
 use core_test_support::responses::ev_function_call;
 use core_test_support::responses::ev_response_created;
+use core_test_support::responses::ev_shell_command_call;
 use core_test_support::responses::mount_models_once;
+use core_test_support::responses::mount_models_once_with_etag;
+use core_test_support::responses::mount_response_once_match;
 use core_test_support::responses::mount_sse_once;
+use core_test_support::responses::mount_sse_once_match;
 use core_test_support::responses::mount_sse_sequence;
 use core_test_support::responses::sse;
 use core_test_support::skip_if_no_network;
@@ -42,6 +46,7 @@ use core_test_support::skip_if_sandbox;
 use core_test_support::wait_for_event;
 use core_test_support::wait_for_event_match;
 use pretty_assertions::assert_eq;
+use serde_json::Value;
 use serde_json::json;
 use tempfile::TempDir;
 use tokio::time::Duration;
@@ -49,9 +54,92 @@ use tokio::time::Instant;
 use tokio::time::sleep;
 use wiremock::BodyPrintLimit;
 use wiremock::MockServer;
+use wiremock::ResponseTemplate;
 
 const REMOTE_MODEL_SLUG: &str = "codex-test";
 
+#[derive(Clone, Default)]
+struct ResponsesMatch {
+    etag: Option<String>,
+    user_text: Option<String>,
+    call_id: Option<String>,
+}
+
+impl ResponsesMatch {
+    fn with_etag(mut self, etag: &str) -> Self {
+        self.etag = Some(etag.to_string());
+        self
+    }
+
+    fn with_user_text(mut self, text: &str) -> Self {
+        self.user_text = Some(text.to_string());
+        self
+    }
+
+    fn with_function_call_output(mut self, call_id: &str) -> Self {
+        self.call_id = Some(call_id.to_string());
+        self
+    }
+}
+
+impl wiremock::Match for ResponsesMatch {
+    fn matches(&self, request: &wiremock::Request) -> bool {
+        if let Some(expected_etag) = &self.etag {
+            let header = request
+                .headers
+                .get("X-If-Models-Match")
+                .and_then(|value| value.to_str().ok());
+            if header != Some(expected_etag.as_str()) {
+                return false;
+            }
+        }
+
+        let Ok(body): Result<Value, _> = request.body_json() else {
+            return false;
+        };
+        let Some(items) = body.get("input").and_then(Value::as_array) else {
+            return false;
+        };
+
+        if let Some(expected_text) = &self.user_text
+            && !input_has_user_text(items, expected_text)
+        {
+            return false;
+        }
+
+        if let Some(expected_call_id) = &self.call_id
+            && !input_has_function_call_output(items, expected_call_id)
+        {
+            return false;
+        }
+
+        true
+    }
+}
+
+fn input_has_user_text(items: &[Value], expected: &str) -> bool {
+    items.iter().any(|item| {
+        item.get("type").and_then(Value::as_str) == Some("message")
+            && item.get("role").and_then(Value::as_str) == Some("user")
+            && item
+                .get("content")
+                .and_then(Value::as_array)
+                .is_some_and(|content| {
+                    content.iter().any(|span| {
+                        span.get("type").and_then(Value::as_str) == Some("input_text")
+                            && span.get("text").and_then(Value::as_str) == Some(expected)
+                    })
+                })
+    })
+}
+
+fn input_has_function_call_output(items: &[Value], call_id: &str) -> bool {
+    items.iter().any(|item| {
+        item.get("type").and_then(Value::as_str) == Some("function_call_output")
+            && item.get("call_id").and_then(Value::as_str) == Some(call_id)
+    })
+}
+
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn remote_models_remote_model_uses_unified_exec() -> Result<()> {
     skip_if_no_network!(Ok(()));
@@ -93,7 +181,6 @@ async fn remote_models_remote_model_uses_unified_exec() -> Result<()> {
         &server,
         ModelsResponse {
             models: vec![remote_model],
-            etag: String::new(),
         },
     )
     .await;
@@ -232,7 +319,6 @@ async fn remote_models_apply_remote_base_instructions() -> Result<()> {
         &server,
         ModelsResponse {
             models: vec![remote_model],
-            etag: String::new(),
         },
     )
     .await;
@@ -299,6 +385,208 @@ async fn remote_models_apply_remote_base_instructions() -> Result<()> {
     Ok(())
 }
 
+/// Exercises the remote-models retry flow:
+/// 1) initial `/models` fetch stores an ETag,
+/// 2) `/responses` uses that ETag for a tool call,
+/// 3) the tool-output turn receives a 412 (stale models),
+/// 4) Codex refreshes `/models` to get a new ETag and retries,
+/// 5) subsequent user turns keep sending the refreshed ETag.
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn remote_models_refresh_etag_after_outdated_models() -> Result<()> {
+    skip_if_no_network!(Ok(()));
+    skip_if_sandbox!(Ok(()));
+
+    let server = MockServer::builder()
+        .body_print_limit(BodyPrintLimit::Limited(80_000))
+        .start()
+        .await;
+
+    let remote_model = test_remote_model("remote-etag", ModelVisibility::List, 1);
+    let initial_etag = "models-etag-initial";
+    let refreshed_etag = "models-etag-refreshed";
+
+    // Phase 1a: seed the initial `/models` response with an ETag.
+    let models_mock = mount_models_once_with_etag(
+        &server,
+        ModelsResponse {
+            models: vec![remote_model.clone()],
+        },
+        initial_etag,
+    )
+    .await;
+
+    // Phase 1b: boot a Codex session configured for remote models.
+    let harness = build_remote_models_harness(&server, |config| {
+        config.features.enable(Feature::RemoteModels);
+        config.model = Some("gpt-5.1".to_string());
+    })
+    .await?;
+
+    let RemoteModelsHarness {
+        codex,
+        cwd,
+        config,
+        conversation_manager,
+        ..
+    } = harness;
+
+    let models_manager = conversation_manager.get_models_manager();
+    wait_for_model_available(&models_manager, "remote-etag", &config).await;
+
+    // Phase 1c: confirm the ETag is stored and `/models` was called.
+    assert_eq!(
+        models_manager.get_models_etag().await.as_deref(),
+        Some(initial_etag),
+    );
+    assert_eq!(
+        models_mock.requests().len(),
+        1,
+        "expected an initial /models request",
+    );
+    assert_eq!(models_mock.requests()[0].url.path(), "/v1/models");
+
+    // Phase 2a: reset mocks so the next `/models` call must be explicit.
+    server.reset().await;
+    // Phase 2b: mount a refreshed `/models` response with a new ETag.
+    let refreshed_models_mock = mount_models_once_with_etag(
+        &server,
+        ModelsResponse {
+            models: vec![remote_model],
+        },
+        refreshed_etag,
+    )
+    .await;
+
+    let call_id = "shell-command-call";
+    let first_prompt = "run a shell command";
+    let followup_prompt = "send another message";
+
+    // Phase 2c: first `/responses` turn uses the initial ETag and emits a tool call.
+    let first_response = mount_sse_once_match(
+        &server,
+        ResponsesMatch::default()
+            .with_etag(initial_etag)
+            .with_user_text(first_prompt),
+        sse(vec![
+            ev_response_created("resp-1"),
+            ev_shell_command_call(call_id, "echo refreshed"),
+            ev_completed("resp-1"),
+        ]),
+    )
+    .await;
+
+    // Phase 2d: the tool-output follow-up returns 412 (stale models).
+    let stale_response = mount_response_once_match(
+        &server,
+        ResponsesMatch::default()
+            .with_etag(initial_etag)
+            .with_function_call_output(call_id),
+        ResponseTemplate::new(412)
+            .set_body_string("Models catalog has changed. Please refresh your models list."),
+    )
+    .await;
+
+    // Phase 2e: retry tool-output follow-up should use the refreshed ETag.
+    let refreshed_response = mount_sse_once_match(
+        &server,
+        ResponsesMatch::default()
+            .with_etag(refreshed_etag)
+            .with_function_call_output(call_id),
+        sse(vec![
+            ev_response_created("resp-2"),
+            ev_assistant_message("msg-1", "done"),
+            ev_completed("resp-2"),
+        ]),
+    )
+    .await;
+
+    // Phase 3a: next user turn should also use the refreshed ETag.
+    let next_turn_response = mount_sse_once_match(
+        &server,
+        ResponsesMatch::default()
+            .with_etag(refreshed_etag)
+            .with_user_text(followup_prompt),
+        sse(vec![
+            ev_response_created("resp-3"),
+            ev_assistant_message("msg-2", "ok"),
+            ev_completed("resp-3"),
+        ]),
+    )
+    .await;
+
+    // Phase 3b: run the first user turn and let retries complete.
+    codex
+        .submit(Op::UserTurn {
+            items: vec![UserInput::Text {
+                text: first_prompt.into(),
+            }],
+            final_output_json_schema: None,
+            cwd: cwd.path().to_path_buf(),
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::DangerFullAccess,
+            model: "gpt-5.1".to_string(),
+            effort: None,
+            summary: ReasoningSummary::Auto,
+        })
+        .await?;
+
+    wait_for_event(&codex, |event| matches!(event, EventMsg::TaskComplete(_))).await;
+
+    // Phase 3c: assert the refresh happened and the ETag was updated.
+    assert_eq!(
+        refreshed_models_mock.requests().len(),
+        1,
+        "expected a refreshed /models request",
+    );
+    assert_eq!(
+        models_manager.get_models_etag().await.as_deref(),
+        Some(refreshed_etag),
+    );
+
+    // Phase 3d: assert the ETag header progression across the retry sequence.
+    assert_eq!(
+        first_response.single_request().header("X-If-Models-Match"),
+        Some(initial_etag.to_string()),
+    );
+    assert_eq!(
+        stale_response.single_request().header("X-If-Models-Match"),
+        Some(initial_etag.to_string()),
+    );
+    assert_eq!(
+        refreshed_response
+            .single_request()
+            .header("X-If-Models-Match"),
+        Some(refreshed_etag.to_string()),
+    );
+
+    // Phase 3e: execute a new user turn and ensure the refreshed ETag persists.
+    codex
+        .submit(Op::UserTurn {
+            items: vec![UserInput::Text {
+                text: followup_prompt.into(),
+            }],
+            final_output_json_schema: None,
+            cwd: cwd.path().to_path_buf(),
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::DangerFullAccess,
+            model: "gpt-5.1".to_string(),
+            effort: None,
+            summary: ReasoningSummary::Auto,
+        })
+        .await?;
+
+    wait_for_event(&codex, |event| matches!(event, EventMsg::TaskComplete(_))).await;
+
+    assert_eq!(
+        next_turn_response
+            .single_request()
+            .header("X-If-Models-Match"),
+        Some(refreshed_etag.to_string()),
+    );
+
+    Ok(())
+}
+
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn remote_models_preserve_builtin_presets() -> Result<()> {
     skip_if_no_network!(Ok(()));
@@ -310,7 +598,6 @@ async fn remote_models_preserve_builtin_presets() -> Result<()> {
         &server,
         ModelsResponse {
             models: vec![remote_model.clone()],
-            etag: String::new(),
         },
     )
     .await;
@@ -330,7 +617,7 @@ async fn remote_models_preserve_builtin_presets() -> Result<()> {
     );
 
     manager
-        .refresh_available_models(&config)
+        .try_refresh_available_models(&config)
         .await
         .expect("refresh succeeds");
 
@@ -368,7 +655,6 @@ async fn remote_models_hide_picker_only_models() -> Result<()> {
         &server,
         ModelsResponse {
             models: vec![remote_model],
-            etag: String::new(),
         },
     )
     .await;
diff --git a/codex-rs/protocol/src/openai_models.rs b/codex-rs/protocol/src/openai_models.rs
index 28b25bb604e..4cdc0e4ba6f 100644
--- a/codex-rs/protocol/src/openai_models.rs
+++ b/codex-rs/protocol/src/openai_models.rs
@@ -197,8 +197,6 @@ pub struct ModelInfo {
 #[derive(Debug, Serialize, Deserialize, Clone, PartialEq, Eq, TS, JsonSchema, Default)]
 pub struct ModelsResponse {
     pub models: Vec<ModelInfo>,
-    #[serde(default)]
-    pub etag: String,
 }
 
 // convert ModelInfo to ModelPreset