From 2d814cd652a63e267c80ae6cae46c31cf6e2106d Mon Sep 17 00:00:00 2001
From: marius-kilocode <marius@kilocode.ai>
Date: Fri, 3 Jul 2026 14:02:53 +0200
Subject: [PATCH] fix(cli): expose off variant and add kilo gateway branch for
 GLM 5.2

GLM 5.2's reasoning variant mapping in ProviderTransform.variants had two
gaps introduced by PR #11555, which adapted upstream commit 22cc758b1a:

1. The @kilocode/kilo-gateway package was not covered, so the gateway fell
   through to the generic GLM binary toggle (instant/thinking) instead of
   GLM 5.2's native effort levels. This made the Kilo gateway show only two
   reasoning efforts while the cloud already pushes none/high/xhigh via
   REASONING_VARIANTS_NONE_HIGH_XHIGH.

2. The off (none) variant was dropped from the openrouter, openai-compatible,
   and anthropic branches, leaving each at only two variants and regressing
   custom openai-compatible providers (e.g. wafer) that previously exposed
   three.

This mirrors the cloud's three-variant definition (none/high/xhigh for the
gateway and openrouter; none/high/max for openai-compatible and anthropic,
matching z.ai native naming) and adds a kilo-gateway case so the computed
fallback matches the config the gateway pushes.
---
 .changeset/fix-glm-52-reasoning-variants.md   |  5 +++++
 packages/opencode/src/provider/transform.ts   | 14 ++++++++++++
 .../opencode/test/provider/transform.test.ts  | 22 +++++++++++++++++++
 3 files changed, 41 insertions(+)
 create mode 100644 .changeset/fix-glm-52-reasoning-variants.md

diff --git a/.changeset/fix-glm-52-reasoning-variants.md b/.changeset/fix-glm-52-reasoning-variants.md
new file mode 100644
index 0000000000..a27fa9758f
--- /dev/null
+++ b/.changeset/fix-glm-52-reasoning-variants.md
@@ -0,0 +1,5 @@
+---
+"kilo-code": patch
+---
+
+Restore the off reasoning variant for GLM 5.2 and add the missing Kilo gateway branch so the effort selector shows all three supported levels (off/high/xhigh) instead of only two.
diff --git a/packages/opencode/src/provider/transform.ts b/packages/opencode/src/provider/transform.ts
index f661d9d633..5844de50eb 100644
--- a/packages/opencode/src/provider/transform.ts
+++ b/packages/opencode/src/provider/transform.ts
@@ -669,21 +669,35 @@ export function variants(model: Provider.Model): Record<string, Record<string, a
   }
   const adaptiveOpus = anthropicOpus47OrLater(model.api.id)
   const adaptiveEfforts = anthropicAdaptiveEfforts(model.api.id)
+  if (glm52 && model.api.npm === "@kilocode/kilo-gateway") {
+    // kilocode_change - mirror the cloud's none/high/xhigh GLM-5.2 variants
+    // (see REASONING_VARIANTS_NONE_HIGH_XHIGH in cloud model-settings.ts) so the
+    // computed fallback matches the config the gateway pushes. Without this the
+    // generic GLM binary toggle (instant/thinking) below is used instead.
+    return {
+      none: { reasoning: { enabled: false, effort: "none" } },
+      high: { reasoning: { enabled: true, effort: "high" } },
+      xhigh: { reasoning: { enabled: true, effort: "xhigh" } },
+    }
+  }
   if (glm52 && model.api.npm === "@openrouter/ai-sdk-provider") {
     // OpenRouter maps xhigh to GLM-5.2's native max effort.
     return {
+      none: { reasoning: { enabled: false } },
       high: { reasoning: { effort: "high" } },
       xhigh: { reasoning: { effort: "xhigh" } },
     }
   }
   if (glm52 && model.api.npm === "@ai-sdk/openai-compatible") {
     return {
+      none: { reasoningEffort: "none" },
       high: { reasoningEffort: "high" },
       max: { reasoningEffort: "max" },
     }
   }
   if (glm52 && model.api.npm === "@ai-sdk/anthropic") {
     return {
+      none: { thinking: { type: "disabled" } },
       high: { effort: "high" },
       max: { effort: "max" },
     }
diff --git a/packages/opencode/test/provider/transform.test.ts b/packages/opencode/test/provider/transform.test.ts
index 709d8dac3a..06583e8988 100644
--- a/packages/opencode/test/provider/transform.test.ts
+++ b/packages/opencode/test/provider/transform.test.ts
@@ -2539,6 +2539,7 @@ describe("ProviderTransform.variants", () => {
       },
     })
     expect(ProviderTransform.variants(model)).toEqual({
+      none: { reasoningEffort: "none" },
       high: { reasoningEffort: "high" },
       max: { reasoningEffort: "max" },
     })
@@ -2555,6 +2556,7 @@ describe("ProviderTransform.variants", () => {
         },
       })
       expect(ProviderTransform.variants(model)).toEqual({
+        none: { reasoningEffort: "none" },
         high: { reasoningEffort: "high" },
         max: { reasoningEffort: "max" },
       })
@@ -2571,6 +2573,7 @@ describe("ProviderTransform.variants", () => {
       },
     })
     expect(ProviderTransform.variants(model)).toEqual({
+      none: { reasoningEffort: "none" },
       high: { reasoningEffort: "high" },
       max: { reasoningEffort: "max" },
     })
@@ -2587,11 +2590,29 @@ describe("ProviderTransform.variants", () => {
       },
     })
     expect(ProviderTransform.variants(model)).toEqual({
+      none: { reasoning: { enabled: false } },
       high: { reasoning: { effort: "high" } },
       xhigh: { reasoning: { effort: "xhigh" } },
     })
   })
 
+  test("glm-5.2 returns effort variants for the kilo gateway", () => {
+    const model = createMockModel({
+      id: "kilo/z-ai/glm-5.2",
+      providerID: "kilo",
+      api: {
+        id: "z-ai/glm-5.2",
+        url: "https://gateway.kilo.ai",
+        npm: "@kilocode/kilo-gateway",
+      },
+    })
+    expect(ProviderTransform.variants(model)).toEqual({
+      none: { reasoning: { enabled: false, effort: "none" } },
+      high: { reasoning: { enabled: true, effort: "high" } },
+      xhigh: { reasoning: { enabled: true, effort: "xhigh" } },
+    })
+  })
+
   test("glm-5.2 returns effort variants for anthropic-compatible providers", () => {
     const model = createMockModel({
       id: "zai-coding-plan/glm-5.2",
@@ -2603,6 +2624,7 @@ describe("ProviderTransform.variants", () => {
       },
     })
     expect(ProviderTransform.variants(model)).toEqual({
+      none: { thinking: { type: "disabled" } },
       high: { effort: "high" },
       max: { effort: "max" },
     })