From 6218b20bf9474fb209dc37279f4d3a92c2f137fe Mon Sep 17 00:00:00 2001
From: redshift <213178690+1ftredsh@users.noreply.github.com>
Date: Mon, 7 Sep 2026 14:11:52 +0200
Subject: [PATCH] revert: use max_tokens for chat/completions instead of
 max_completion_tokens

Revert the default request-limit field back to legacy max_tokens for
chat/completions (keeping max_output_tokens for the Responses API).
---
 src/daemon/http/index.ts | 11 +++--------
 1 file changed, 3 insertions(+), 8 deletions(-)

diff --git a/src/daemon/http/index.ts b/src/daemon/http/index.ts
index 354d42f..505b6c8 100644
--- a/src/daemon/http/index.ts
+++ b/src/daemon/http/index.ts
@@ -1712,20 +1712,15 @@ export function createDaemonRequestHandler(deps: {
     // limit. Without this, the SDK prices at the provider's worst-case
     // max_completion_cost, which varies widely across providers (2.3× for
     // kimi-k3) and balloons during provider failover. Chat/completions use
-    // max_completion_tokens (the OpenAI-standard field; legacy max_tokens is
-    // deprecated and rejected by reasoning models); the Responses API uses
-    // max_output_tokens.
+    // max_tokens; the OpenAI Responses API uses max_output_tokens.
     if (deps.maxTokens > 0) {
       const isResponsesPath = url.pathname.includes("/responses");
       if (isResponsesPath) {
         if (typeof bodyObj.max_output_tokens !== "number") {
           bodyObj.max_output_tokens = deps.maxTokens;
         }
-      } else if (
-        typeof bodyObj.max_completion_tokens !== "number" &&
-        typeof bodyObj.max_tokens !== "number"
-      ) {
-        bodyObj.max_completion_tokens = deps.maxTokens;
+      } else if (typeof bodyObj.max_tokens !== "number") {
+        bodyObj.max_tokens = deps.maxTokens;
       }
     }