From 6712cc4c2c2be26c5baa2db9af6c530ca950b713 Mon Sep 17 00:00:00 2001
From: Aiden Cline <63023139+rekram1-node@users.noreply.github.com>
Date: Tue, 29 Sep 2026 00:45:17 -0500
Subject: [PATCH] fix(core): turn off Workers AI thinking through the chat
 template (#52017)

---
 packages/core/src/variant.ts       | 21 +++++++++++++++++++-
 packages/core/test/variant.test.ts | 31 ++++++++++++++++++++++++++++++
 2 files changed, 51 insertions(+), 1 deletion(-)

diff --git a/packages/core/src/variant.ts b/packages/core/src/variant.ts
index 1f47d37bf41..c07fcf13cee 100644
--- a/packages/core/src/variant.ts
+++ b/packages/core/src/variant.ts
@@ -160,6 +160,25 @@ const nvidiaChat: Protocol = (model, support) => {
   }
 }
 
+// Workers AI ignores or rejects `reasoning_effort: "none"` on most models. Kimi's chat template reads
+// `thinking`; the others read `enable_thinking`, so both are sent.
+const workersAITemplate = (thinking: boolean): Overlay => ({
+  body: { chat_template_kwargs: { enable_thinking: thinking, thinking } },
+})
+
+const workersAIChat: Protocol = (_, support) => {
+  switch (support.type) {
+    case "effort":
+      return efforts(support.values ?? EFFORTS, (effort) =>
+        effort === "none" ? workersAITemplate(false) : { settings: { reasoningEffort: effort } },
+      )
+    case "toggle":
+      return toggle(workersAITemplate(false), workersAITemplate(true))
+    case "budget_tokens":
+      return []
+  }
+}
+
 const basetenTemplate = (enable_thinking: boolean): Overlay => ({
   body: { chat_template_args: { enable_thinking } },
 })
@@ -552,7 +571,7 @@ const PROTOCOLS: Readonly<Record<string, Protocol>> = {
   "@opencode/ai/providers/alibaba/chat": alibabaChat,
   "@opencode/ai/providers/baseten": basetenChat,
   "@opencode/ai/providers/cerebras": openaiChat,
-  "@opencode/ai/providers/cloudflare-workers-ai": openaiChat,
+  "@opencode/ai/providers/cloudflare-workers-ai": workersAIChat,
   "@opencode/ai/providers/deepinfra": deepinfraChat,
   "@opencode/ai/providers/deepseek": deepseekChat,
   "@opencode/ai/providers/fireworks": openaiChat,
diff --git a/packages/core/test/variant.test.ts b/packages/core/test/variant.test.ts
index c763b9f17b9..14afa38f265 100644
--- a/packages/core/test/variant.test.ts
+++ b/packages/core/test/variant.test.ts
@@ -264,6 +264,37 @@ test("caps Alibaba thinking budget variants at 64k", () => {
   ])
 })
 
+test("spells Workers AI thinking controls through the chat template", () => {
+  const pkg = "@opencode/ai/providers/cloudflare-workers-ai"
+  const off = { body: { chat_template_kwargs: { enable_thinking: false, thinking: false } } }
+  const on = { body: { chat_template_kwargs: { enable_thinking: true, thinking: true } } }
+  expect(
+    resolve(model(pkg, "@cf/deepseek-ai/deepseek-v4-flash-0731"), [
+      { type: "effort", values: ["none", "low", "high", "max"] },
+    ]),
+  ).toEqual([
+    { id: "none", ...off },
+    { id: "low", settings: { reasoningEffort: "low" } },
+    { id: "high", settings: { reasoningEffort: "high" } },
+    { id: "max", settings: { reasoningEffort: "max" } },
+  ])
+  expect(resolve(model(pkg, "@cf/zai-org/glm-4.7-flash"), [{ type: "toggle" }])).toEqual([
+    { id: "none", ...off },
+    { id: "thinking", ...on },
+  ])
+  expect(
+    resolve(model(pkg, "@cf/qwen/qwen3.8-27b"), [
+      { type: "toggle" },
+      { type: "effort", values: ["low", "medium", "xhigh"] },
+    ]),
+  ).toEqual([
+    { id: "none", ...off },
+    { id: "low", settings: { reasoningEffort: "low" } },
+    { id: "medium", settings: { reasoningEffort: "medium" } },
+    { id: "xhigh", settings: { reasoningEffort: "xhigh" } },
+  ])
+})
+
 test("spells Chat Completions variants for hosting providers", () => {
   expect(
     resolve(model("@opencode/ai/providers/openai-compatible", "deepseek-ai/deepseek-v4-pro", undefined, "nvidia"), [