From 9c70ab93d38e8ceba75128caa7ce1cd1e3b41d42 Mon Sep 17 00:00:00 2001
From: Dizzle <112548150+maxmad64bis@users.noreply.github.com>
Date: Sun, 25 Oct 2026 14:38:55 +0100
Subject: [PATCH] fix(sse): keep explicit thinking type over connection default

explicit thinking type survives the connection default; missing type still receives it
---
 ...m-thinking-type-ecrase-client-explicite.md |  1 +
 open-sse/executors/glm.ts                     | 23 ++++++--
 tests/unit/glm-thinking-type-explicit.test.ts | 54 +++++++++++++++++++
 3 files changed, 73 insertions(+), 5 deletions(-)
 create mode 100644 changelog.d/fixes/18114-glm-thinking-type-ecrase-client-explicite.md
 create mode 100644 tests/unit/glm-thinking-type-explicit.test.ts

diff --git a/changelog.d/fixes/18114-glm-thinking-type-ecrase-client-explicite.md b/changelog.d/fixes/18114-glm-thinking-type-ecrase-client-explicite.md
new file mode 100644
index 000000000000..4486b8971a7d
--- /dev/null
+++ b/changelog.d/fixes/18114-glm-thinking-type-ecrase-client-explicite.md
@@ -0,0 +1 @@
+- **fix(sse):** keep an explicit client thinking type instead of overwriting it with the connection default ([#18114](https://github.com/cheaper-inference/OmniRoute/pull/18114)) — thanks @maxmad64bis
diff --git a/open-sse/executors/glm.ts b/open-sse/executors/glm.ts
index 581849a43ac2..e7c798591ee8 100644
--- a/open-sse/executors/glm.ts
+++ b/open-sse/executors/glm.ts
@@ -129,16 +129,24 @@ function isGlmThinkingModel(model: string): boolean {
  */
 const GLM_THINKING_DEFAULT_MAX_TOKENS = 131072;
 
-function applyGlmRequestDefaults(body: unknown, defaults?: JsonRecord | null): unknown {
+function applyGlmRequestDefaults(
+  body: unknown,
+  defaults?: JsonRecord | null,
+  clientThinkingType?: unknown
+): unknown {
   const record = asRecord(body);
   if (!record || !defaults) return body;
 
   const next = { ...(applyProviderRequestDefaults(record, defaults) as JsonRecord) };
   const thinkingType = typeof defaults.thinkingType === "string" ? defaults.thinkingType : null;
 
-  if (thinkingType && next.thinking === undefined) {
-    next.thinking = { type: thinkingType };
-  } else if (thinkingType && asRecord(next.thinking)?.type === "enabled") {
+  // Apply the connection default only when the client did not set a thinking
+  // type of its own; other thinking fields are merged and preserved.
+  if (
+    thinkingType &&
+    clientThinkingType === undefined &&
+    (next.thinking === undefined || asRecord(next.thinking))
+  ) {
     next.thinking = { ...asRecord(next.thinking), type: thinkingType };
   }
 
@@ -310,8 +318,13 @@ export class GlmExecutor extends DefaultExecutor {
     stream: boolean,
     credentials: ProviderCredentials
   ) {
+    const clientThinkingType = asRecord(asRecord(body)?.thinking)?.type;
     const cleanedBody = super.transformRequest(model, body, stream, credentials);
-    return applyGlmRequestDefaults(cleanedBody, this.config.requestDefaults as JsonRecord | null);
+    return applyGlmRequestDefaults(
+      cleanedBody,
+      this.config.requestDefaults as JsonRecord | null,
+      clientThinkingType
+    );
   }
 
   transformForTransport(
diff --git a/tests/unit/glm-thinking-type-explicit.test.ts b/tests/unit/glm-thinking-type-explicit.test.ts
new file mode 100644
index 000000000000..e5494b0461b6
--- /dev/null
+++ b/tests/unit/glm-thinking-type-explicit.test.ts
@@ -0,0 +1,54 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import { GlmExecutor } from "../../open-sse/executors/glm.ts";
+
+type ThinkingBody = { thinking?: unknown };
+
+test("GlmExecutor keeps explicit enabled thinking type over adaptive default", () => {
+  const executor = new GlmExecutor("glmt");
+  const body = {
+    messages: [{ role: "user", content: "hi" }],
+    thinking: { type: "enabled", budget_tokens: 1000 },
+  };
+
+  const transformed = executor.transformRequest("glm-5.1", body, true, {
+    apiKey: "glm-key",
+  }) as ThinkingBody;
+
+  assert.deepEqual(transformed.thinking, { type: "enabled", budget_tokens: 1000 });
+});
+
+test("GlmExecutor applies adaptive default when thinking type is missing", () => {
+  const executor = new GlmExecutor("glmt");
+
+  const withoutThinking = executor.transformRequest(
+    "glm-5.1",
+    { messages: [{ role: "user", content: "hi" }] },
+    true,
+    { apiKey: "glm-key" }
+  ) as ThinkingBody;
+  assert.deepEqual(withoutThinking.thinking, { type: "adaptive", budget_tokens: 24_576 });
+
+  const typeless = executor.transformRequest(
+    "glm-5.1",
+    { messages: [{ role: "user", content: "hi" }], thinking: { budget_tokens: 1000 } },
+    true,
+    { apiKey: "glm-key" }
+  ) as ThinkingBody;
+  assert.deepEqual(typeless.thinking, { type: "adaptive", budget_tokens: 1000 });
+});
+
+test("GlmExecutor leaves thinking untouched without connection default", () => {
+  const executor = new GlmExecutor("glm");
+  const body = {
+    messages: [{ role: "user", content: "hi" }],
+    thinking: { type: "enabled", budget_tokens: 1000 },
+  };
+
+  const transformed = executor.transformRequest("glm-5.1", body, false, {
+    apiKey: "glm-key",
+  }) as ThinkingBody;
+
+  assert.deepEqual(transformed.thinking, { type: "enabled", budget_tokens: 1000 });
+});