diff --git a/.changeset/drop-max-tokens-gpt5-openai-compatible.md b/.changeset/drop-max-tokens-gpt5-openai-compatible.md
new file mode 100644
index 00000000000..1bfdbdfa9b3
--- /dev/null
+++ b/.changeset/drop-max-tokens-gpt5-openai-compatible.md
@@ -0,0 +1,5 @@
+---
+"@kilocode/cli": patch
+---
+
+Fix gpt-5 models routed through OpenAI-compatible providers (e.g. LiteLLM) rejecting requests with "Unsupported parameter: max_tokens". The CLI now drops the max output token cap for gpt-5 models on `@ai-sdk/openai-compatible`, letting the upstream default output budget apply.
diff --git a/packages/opencode/src/kilocode/plugin/drop-max-tokens.ts b/packages/opencode/src/kilocode/plugin/drop-max-tokens.ts
new file mode 100644
index 00000000000..404ccd02db0
--- /dev/null
+++ b/packages/opencode/src/kilocode/plugin/drop-max-tokens.ts
@@ -0,0 +1,17 @@
+import type { Hooks, PluginInput } from "@kilocode/plugin"
+
+// Drop the max output tokens cap for gpt-5 models routed through
+// @ai-sdk/openai-compatible. The SDK always emits `max_tokens`, but OpenAI's
+// gpt-5 family (and compatible proxies like LiteLLM) rejects that field and
+// requires `max_completion_tokens` instead. The openai-compatible SDK has no
+// way to rename the field, so we clear the cap and let the upstream default
+// output budget apply.
+export async function DropMaxTokensForGpt5Plugin(_input: PluginInput): Promise<Hooks> {
+  return {
+    "chat.params": async (input, output) => {
+      if (input.model.api.npm !== "@ai-sdk/openai-compatible") return
+      if (!input.model.api.id.toLowerCase().includes("gpt-5")) return
+      output.maxOutputTokens = undefined
+    },
+  }
+}
diff --git a/packages/opencode/src/plugin/index.ts b/packages/opencode/src/plugin/index.ts
index e944b2f4f95..e796264fb44 100644
--- a/packages/opencode/src/plugin/index.ts
+++ b/packages/opencode/src/plugin/index.ts
@@ -25,6 +25,7 @@ import { errorMessage } from "@/util/error"
 import { PluginLoader } from "./loader"
 import { parsePluginSpecifier, readPluginId, readV1Plugin, resolvePluginId } from "./shared"
 import { KiloAuthPlugin } from "@kilocode/kilo-gateway" // kilocode_change
+import { DropMaxTokensForGpt5Plugin } from "@/kilocode/plugin/drop-max-tokens" // kilocode_change
 import { registerAdapter } from "@/control-plane/adapters"
 import type { WorkspaceAdapter } from "@/control-plane/types"
 
@@ -67,6 +68,7 @@ const INTERNAL_PLUGINS: PluginInstance[] = [
   CloudflareWorkersAuthPlugin,
   CloudflareAIGatewayAuthPlugin,
   AzureAuthPlugin,
+  DropMaxTokensForGpt5Plugin,
 ]
 // kilocode_change end
 
diff --git a/packages/opencode/test/kilocode/drop-max-tokens-gpt5.test.ts b/packages/opencode/test/kilocode/drop-max-tokens-gpt5.test.ts
new file mode 100644
index 00000000000..9c6c677b55b
--- /dev/null
+++ b/packages/opencode/test/kilocode/drop-max-tokens-gpt5.test.ts
@@ -0,0 +1,72 @@
+import { expect, test } from "bun:test"
+import { DropMaxTokensForGpt5Plugin } from "@/kilocode/plugin/drop-max-tokens"
+
+const pluginInput = {
+  client: {} as never,
+  project: {} as never,
+  directory: "",
+  worktree: "",
+  experimental_workspace: { register() {} },
+  serverUrl: new URL("https://example.com"),
+  $: {} as never,
+}
+
+function makeHookInput(overrides: { providerID?: string; apiId?: string; npm?: string; reasoning?: boolean }) {
+  return {
+    sessionID: "s",
+    agent: "a",
+    provider: {} as never,
+    message: {} as never,
+    model: {
+      providerID: overrides.providerID ?? "litellm",
+      api: {
+        id: overrides.apiId ?? "gpt-5",
+        url: "",
+        npm: overrides.npm ?? "@ai-sdk/openai-compatible",
+      },
+      capabilities: {
+        reasoning: overrides.reasoning ?? true,
+        temperature: false,
+        attachment: true,
+        toolcall: true,
+        input: { text: true, audio: false, image: false, video: false, pdf: false },
+        output: { text: true, audio: false, image: false, video: false, pdf: false },
+        interleaved: false,
+      },
+    } as never,
+  }
+}
+
+function makeHookOutput() {
+  return { temperature: 0, topP: 1, topK: 0, maxOutputTokens: 32_000 as number | undefined, options: {} }
+}
+
+test("drops maxOutputTokens for gpt-5 on @ai-sdk/openai-compatible", async () => {
+  const hooks = await DropMaxTokensForGpt5Plugin(pluginInput)
+  const out = makeHookOutput()
+  await hooks["chat.params"]!(makeHookInput({ apiId: "gpt-5" }), out)
+  expect(out.maxOutputTokens).toBeUndefined()
+})
+
+test("drops maxOutputTokens for gpt-5 variants (codex, pro, mini) on openai-compatible", async () => {
+  const hooks = await DropMaxTokensForGpt5Plugin(pluginInput)
+  for (const id of ["gpt-5.2-codex", "gpt-5-pro", "gpt-5-mini", "openai/gpt-5-turbo"]) {
+    const out = makeHookOutput()
+    await hooks["chat.params"]!(makeHookInput({ apiId: id }), out)
+    expect(out.maxOutputTokens).toBeUndefined()
+  }
+})
+
+test("keeps maxOutputTokens for gpt-5 on non-openai-compatible SDKs", async () => {
+  const hooks = await DropMaxTokensForGpt5Plugin(pluginInput)
+  const out = makeHookOutput()
+  await hooks["chat.params"]!(makeHookInput({ apiId: "gpt-5", npm: "@ai-sdk/openai" }), out)
+  expect(out.maxOutputTokens).toBe(32_000)
+})
+
+test("keeps maxOutputTokens for non-gpt-5 models on openai-compatible", async () => {
+  const hooks = await DropMaxTokensForGpt5Plugin(pluginInput)
+  const out = makeHookOutput()
+  await hooks["chat.params"]!(makeHookInput({ apiId: "gpt-4-turbo" }), out)
+  expect(out.maxOutputTokens).toBe(32_000)
+})