From f9178e947e36a52bf65678b546b1ec0701d0d78f Mon Sep 17 00:00:00 2001
From: "Rodrigo Rodriguez (Pragmatismo)" <me@rodrigorodriguez.com>
Date: Thu, 16 Apr 2026 16:21:16 -0300
Subject: [PATCH] Use max_completion_tokens for GPT-5 models

---
 src/llm/mod.rs | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/src/llm/mod.rs b/src/llm/mod.rs
index 012db11e..0f5967c6 100644
--- a/src/llm/mod.rs
+++ b/src/llm/mod.rs
@@ -506,11 +506,17 @@ impl LLMProvider for OpenAIClient {
         }
 
         // Build the request body - include tools if provided
+        // GPT-5 models use max_completion_tokens instead of max_tokens
+        let token_key = if model.contains("gpt-5") {
+            "max_completion_tokens"
+        } else {
+            "max_tokens"
+        };
         let mut request_body = serde_json::json!({
             "model": model,
             "messages": messages,
             "stream": true,
-            "max_tokens": 16384,
+            token_key: 16384,
             "temperature": 1.0,
             "top_p": 1.0
         });