From da4a91b3647f90fcab3c980150e6d0d476bdc28d Mon Sep 17 00:00:00 2001
From: Ryan Loney <ryan.loney@cerebras.net>
Date: Mon, 24 Aug 2026 04:34:00 -0700
Subject: [PATCH] fix(opencode): preserve Cerebras completion limit (#13289)

* fix(opencode): preserve Cerebras completion limit

* chore: annotate Kilo-specific changes
---
 packages/opencode/src/provider/transform.ts   | 13 ++++++
 packages/opencode/src/session/llm.ts          | 16 ++++++-
 .../opencode/test/provider/transform.test.ts  | 42 +++++++++++++++++++
 3 files changed, 69 insertions(+), 2 deletions(-)

diff --git a/packages/opencode/src/provider/transform.ts b/packages/opencode/src/provider/transform.ts
index 386a174741..a2c6094126 100644
--- a/packages/opencode/src/provider/transform.ts
+++ b/packages/opencode/src/provider/transform.ts
@@ -1724,6 +1724,19 @@ export function maxOutputTokens(model: Provider.Model, outputTokenMax = OUTPUT_T
   return Math.min(model.limit.output, outputTokenMax) || outputTokenMax
 }
 
+// kilocode_change start
+export function maxOutputTokensForRequest(input: {
+  model: Provider.Model
+  options: Record<string, any>
+  maxOutputTokens: number | undefined
+}): number | undefined {
+  if (input.model.api.npm === "@ai-sdk/cerebras" && input.options.max_completion_tokens !== undefined) {
+    return undefined
+  }
+  return input.maxOutputTokens
+}
+// kilocode_change end
+
 type JsonRecord = Record<string, unknown>
 
 function isPlainObject(value: unknown): value is JsonRecord {
diff --git a/packages/opencode/src/session/llm.ts b/packages/opencode/src/session/llm.ts
index 9f16e2e017..620a8e6fe2 100644
--- a/packages/opencode/src/session/llm.ts
+++ b/packages/opencode/src/session/llm.ts
@@ -312,7 +312,13 @@ const live: Layer.Layer<
           temperature: prepared.params.temperature,
           topP: prepared.params.topP,
           topK: prepared.params.topK,
-          maxOutputTokens: prepared.params.maxOutputTokens,
+          // kilocode_change start
+          maxOutputTokens: ProviderTransform.maxOutputTokensForRequest({
+            model: input.model,
+            options: prepared.params.options,
+            maxOutputTokens: prepared.params.maxOutputTokens,
+          }),
+          // kilocode_change end
           providerOptions: prepared.params.options,
           headers: prepared.headers,
           abort: input.abort,
@@ -388,9 +394,15 @@ const live: Layer.Layer<
         topK: prepared.params.topK,
         providerOptions: ProviderTransform.providerOptions(input.model, prepared.params.options),
         activeTools: Object.keys(prepared.tools).filter((x) => x !== "invalid"),
+        // kilocode_change start
         tools: prepared.tools,
         toolChoice: input.toolChoice,
-        maxOutputTokens: prepared.params.maxOutputTokens,
+        maxOutputTokens: ProviderTransform.maxOutputTokensForRequest({
+          model: input.model,
+          options: prepared.params.options,
+          maxOutputTokens: prepared.params.maxOutputTokens,
+        }),
+        // kilocode_change end
         abortSignal: input.abort,
         ...KiloLLM.timeout({ options: prepared.params.options, fallback: item.options, log: l }), // kilocode_change
         headers: prepared.headers,
diff --git a/packages/opencode/test/provider/transform.test.ts b/packages/opencode/test/provider/transform.test.ts
index 0a96b4782a..f5107fabdf 100644
--- a/packages/opencode/test/provider/transform.test.ts
+++ b/packages/opencode/test/provider/transform.test.ts
@@ -4699,6 +4699,48 @@ describe("ProviderTransform.variants", () => {
       expect(result.low).toEqual({ reasoningEffort: "low" })
       expect(result.high).toEqual({ reasoningEffort: "high" })
     })
+
+    // kilocode_change start
+    test("omits the generic output cap when an exact max_completion_tokens value is configured", () => {
+      const model = createMockModel({
+        id: "cerebras/gpt-oss-120b",
+        providerID: "cerebras",
+        api: {
+          id: "gpt-oss-120b",
+          url: "https://api.cerebras.ai/v1",
+          npm: "@ai-sdk/cerebras",
+        },
+      })
+
+      const result = ProviderTransform.maxOutputTokensForRequest({
+        model,
+        options: { max_completion_tokens: 64 },
+        maxOutputTokens: 32_000,
+      })
+
+      expect(result).toBeUndefined()
+    })
+
+    test("keeps the generic output cap when no exact value is configured", () => {
+      const model = createMockModel({
+        id: "cerebras/gpt-oss-120b",
+        providerID: "cerebras",
+        api: {
+          id: "gpt-oss-120b",
+          url: "https://api.cerebras.ai/v1",
+          npm: "@ai-sdk/cerebras",
+        },
+      })
+
+      const result = ProviderTransform.maxOutputTokensForRequest({
+        model,
+        options: {},
+        maxOutputTokens: 32_000,
+      })
+
+      expect(result).toBe(32_000)
+    })
+    // kilocode_change end
   })
 
   describe("@ai-sdk/togetherai", () => {