diff --git a/crates/tui/src/client.rs b/crates/tui/src/client.rs
index 50998854..b6538458 100644
--- a/crates/tui/src/client.rs
+++ b/crates/tui/src/client.rs
@@ -1265,8 +1265,14 @@ mod tests {
 
     #[test]
     fn generic_openai_provider_drops_deepseek_reasoning_content() {
+        // #1542 intent (narrowed by #1739/#1694): a *genuine non-DeepSeek*
+        // model on the generic openai provider must not carry DeepSeek-only
+        // `reasoning_content`. A DeepSeek reasoning model on the openai
+        // provider (DeepSeek-compatible endpoint) is now covered separately
+        // and DOES replay reasoning_content — see
+        // `deepseek_model_on_openai_provider_still_replays_reasoning_content`.
         let request = MessageRequest {
-            model: "deepseek-v4-pro".to_string(),
+            model: "gpt-4o".to_string(),
             messages: vec![Message {
                 role: "assistant".to_string(),
                 content: vec![
@@ -1291,19 +1297,6 @@ mod tests {
             top_p: None,
         };
 
-        let deepseek =
-            build_chat_messages_for_request_and_provider(&request, ApiProvider::Deepseek);
-        let native_assistant = deepseek
-            .iter()
-            .find(|value| value.get("role").and_then(Value::as_str) == Some("assistant"))
-            .expect("assistant message");
-        assert_eq!(
-            native_assistant
-                .get("reasoning_content")
-                .and_then(Value::as_str),
-            Some("plan")
-        );
-
         let openai = build_chat_messages_for_request_and_provider(&request, ApiProvider::Openai);
         let generic_assistant = openai
             .iter()
@@ -2707,8 +2700,12 @@ mod tests {
 
     #[test]
     fn sanitize_thinking_mode_skips_generic_openai_provider() {
+        // #1542 intent (narrowed by #1739/#1694): the sanitizer only skips for
+        // a *genuine non-DeepSeek* model on the generic openai provider. A
+        // DeepSeek reasoning model on the openai provider still gets sanitized
+        // (see chat.rs `deepseek_model_on_openai_provider_still_replays_*`).
         let mut body = json!({
-            "model": "deepseek-v4-pro",
+            "model": "gpt-4o",
             "messages": [
                 { "role": "user", "content": "hi" },
                 {
@@ -2721,7 +2718,7 @@ mod tests {
 
         let result = sanitize_thinking_mode_messages(
             &mut body,
-            "deepseek-v4-pro",
+            "gpt-4o",
             Some("max"),
             ApiProvider::Openai,
         );
diff --git a/crates/tui/src/client/chat.rs b/crates/tui/src/client/chat.rs
index 8e9af335..15dc3927 100644
--- a/crates/tui/src/client/chat.rs
+++ b/crates/tui/src/client/chat.rs
@@ -1628,7 +1628,12 @@ fn should_replay_reasoning_content_for_provider(
     model: &str,
     effort: Option<&str>,
 ) -> bool {
-    if !provider_accepts_reasoning_content(provider) {
+    if !provider_accepts_reasoning_content(provider) && !requires_reasoning_content(model) {
+        // Generic non-DeepSeek model on a provider that rejects the field:
+        // keep stripping it (preserves the #1542 fix). But a known DeepSeek
+        // reasoning model pointed at a DeepSeek-compatible endpoint via the
+        // generic `openai` provider still requires reasoning_content replay,
+        // or the thinking-mode API returns 400 (#1739 / #1694).
         return false;
     }
     should_replay_reasoning_content(model, effort)
@@ -2828,7 +2833,7 @@ mod alias_thinking_detection_tests {
     //! https://api-docs.deepseek.com/guides/thinking_mode
     use super::{
         provider_accepts_reasoning_content, requires_reasoning_content,
-        should_replay_reasoning_content,
+        should_replay_reasoning_content, should_replay_reasoning_content_for_provider,
     };
     use crate::config::ApiProvider;
 
@@ -2897,4 +2902,49 @@ mod alias_thinking_detection_tests {
         assert!(provider_accepts_reasoning_content(ApiProvider::Deepseek));
         assert!(provider_accepts_reasoning_content(ApiProvider::NvidiaNim));
     }
+
+    #[test]
+    fn deepseek_model_on_openai_provider_still_replays_reasoning_content() {
+        // #1739 / #1694: a DeepSeek thinking model pointed at a
+        // DeepSeek-compatible endpoint via the generic `openai` provider must
+        // still replay reasoning_content, even though the provider itself does
+        // not accept the field. Otherwise the thinking-mode API returns 400.
+        assert!(should_replay_reasoning_content_for_provider(
+            ApiProvider::Openai,
+            "deepseek-v4-flash",
+            None,
+        ));
+        assert!(should_replay_reasoning_content_for_provider(
+            ApiProvider::Openai,
+            "deepseek-v4-pro",
+            None,
+        ));
+        assert!(should_replay_reasoning_content_for_provider(
+            ApiProvider::Openai,
+            "deepseek-reasoner",
+            Some("medium"),
+        ));
+        // The documented escape hatch still wins over model detection.
+        assert!(!should_replay_reasoning_content_for_provider(
+            ApiProvider::Openai,
+            "deepseek-v4-flash",
+            Some("off"),
+        ));
+    }
+
+    #[test]
+    fn generic_model_on_openai_provider_still_strips_reasoning_content() {
+        // #1542 no-regression guard: a genuine non-DeepSeek model on the
+        // openai provider must continue to have reasoning_content stripped.
+        assert!(!should_replay_reasoning_content_for_provider(
+            ApiProvider::Openai,
+            "gpt-4o",
+            None,
+        ));
+        assert!(!should_replay_reasoning_content_for_provider(
+            ApiProvider::Openai,
+            "claude-sonnet-4-6",
+            None,
+        ));
+    }
 }