From f38facd43064b8415053ef098ec3e582805962e5 Mon Sep 17 00:00:00 2001
From: Hemilt0n <akobama233@gmail.com>
Date: Sun, 15 Mar 2026 20:39:17 +0800
Subject: [PATCH] fix(proxy): use max_completion_tokens for o1/o3 series models
 (#1451)

* fix(proxy): use max_completion_tokens for o1/o3 series models

When converting Anthropic requests to OpenAI format for o1/o3 series
models (like o1-mini, o3-mini), use max_completion_tokens instead of
max_tokens to avoid unsupported_parameter errors.

Fixes #1448

* fix: revert incorrect o-series max_completion_tokens in Responses API path

Responses API uses max_output_tokens for all models including o-series.
The o-series max_completion_tokens fix should only apply to Chat Completions API.

---------

Co-authored-by: Hajen Teowideo <hajen.teowideo@example.com>
Co-authored-by: Jason Young <44939412+farion1231@users.noreply.github.com>
Co-authored-by: Jason <farion1231@gmail.com>
---
 src-tauri/src/proxy/providers/transform.rs    | 65 ++++++++++++++++++-
 .../proxy/providers/transform_responses.rs    | 15 ++++-
 2 files changed, 77 insertions(+), 3 deletions(-)

diff --git a/src-tauri/src/proxy/providers/transform.rs b/src-tauri/src/proxy/providers/transform.rs
index 65fa4784e..3eb6f6565 100644
--- a/src-tauri/src/proxy/providers/transform.rs
+++ b/src-tauri/src/proxy/providers/transform.rs
@@ -6,6 +6,14 @@
 use crate::proxy::error::ProxyError;
 use serde_json::{json, Value};
 
+/// Detect OpenAI o-series reasoning models (o1, o3, o4-mini, etc.)
+/// These models require `max_completion_tokens` instead of `max_tokens`.
+pub fn is_openai_o_series(model: &str) -> bool {
+    model.len() > 1
+        && model.starts_with('o')
+        && model.as_bytes().get(1).is_some_and(|b| b.is_ascii_digit())
+}
+
 /// Anthropic 请求 → OpenAI 请求
 ///
 /// `cache_key`: optional prompt_cache_key to inject for improved cache routing
@@ -50,9 +58,14 @@ pub fn anthropic_to_openai(body: Value, cache_key: Option<&str>) -> Result<Value
 
     result["messages"] = json!(messages);
 
-    // 转换参数
+    // 转换参数 — o-series 模型需要 max_completion_tokens
+    let model = body.get("model").and_then(|m| m.as_str()).unwrap_or("");
     if let Some(v) = body.get("max_tokens") {
-        result["max_tokens"] = v.clone();
+        if is_openai_o_series(model) {
+            result["max_completion_tokens"] = v.clone();
+        } else {
+            result["max_tokens"] = v.clone();
+        }
     }
     if let Some(v) = body.get("temperature") {
         result["temperature"] = v.clone();
@@ -772,4 +785,52 @@ mod tests {
         assert_eq!(result["content"][1]["type"], "text");
         assert_eq!(result["content"][1]["text"], "I can't do that");
     }
+
+    #[test]
+    fn test_is_openai_o_series() {
+        assert!(is_openai_o_series("o1"));
+        assert!(is_openai_o_series("o1-preview"));
+        assert!(is_openai_o_series("o1-mini"));
+        assert!(is_openai_o_series("o3"));
+        assert!(is_openai_o_series("o3-mini"));
+        assert!(is_openai_o_series("o4-mini"));
+        assert!(!is_openai_o_series("gpt-4o"));
+        assert!(!is_openai_o_series("openai-gpt"));
+        assert!(!is_openai_o_series("o"));
+        assert!(!is_openai_o_series(""));
+    }
+
+    #[test]
+    fn test_anthropic_to_openai_o_series_max_completion_tokens() {
+        for model in &["o1", "o3-mini", "o4-mini"] {
+            let input = json!({
+                "model": model,
+                "max_tokens": 4096,
+                "messages": [{"role": "user", "content": "Hello"}]
+            });
+
+            let result = anthropic_to_openai(input, None).unwrap();
+            assert!(
+                result.get("max_tokens").is_none(),
+                "{model} should not have max_tokens"
+            );
+            assert_eq!(
+                result["max_completion_tokens"], 4096,
+                "{model} should use max_completion_tokens"
+            );
+        }
+    }
+
+    #[test]
+    fn test_anthropic_to_openai_non_o_series_keeps_max_tokens() {
+        let input = json!({
+            "model": "gpt-4o",
+            "max_tokens": 1024,
+            "messages": [{"role": "user", "content": "Hello"}]
+        });
+
+        let result = anthropic_to_openai(input, None).unwrap();
+        assert_eq!(result["max_tokens"], 1024);
+        assert!(result.get("max_completion_tokens").is_none());
+    }
 }
diff --git a/src-tauri/src/proxy/providers/transform_responses.rs b/src-tauri/src/proxy/providers/transform_responses.rs
index 8f223b725..b11eb9b50 100644
--- a/src-tauri/src/proxy/providers/transform_responses.rs
+++ b/src-tauri/src/proxy/providers/transform_responses.rs
@@ -45,7 +45,7 @@ pub fn anthropic_to_responses(body: Value, cache_key: Option<&str>) -> Result<Va
         result["input"] = json!(input);
     }
 
-    // max_tokens → max_output_tokens
+    // max_tokens → max_output_tokens (Responses API uses max_output_tokens for all models)
     if let Some(v) = body.get("max_tokens") {
         result["max_output_tokens"] = v.clone();
     }
@@ -897,4 +897,17 @@ mod tests {
         assert_eq!(result["usage"]["cache_read_input_tokens"], 60);
         assert_eq!(result["usage"]["cache_creation_input_tokens"], 20);
     }
+
+    #[test]
+    fn test_anthropic_to_responses_o_series_uses_max_output_tokens() {
+        // Responses API always uses max_output_tokens, even for o-series models
+        let input = json!({
+            "model": "o3-mini",
+            "max_tokens": 4096,
+            "messages": [{"role": "user", "content": "Hello"}]
+        });
+        let result = anthropic_to_responses(input, None).unwrap();
+        assert_eq!(result["max_output_tokens"], 4096);
+        assert!(result.get("max_completion_tokens").is_none());
+    }
 }