fix(opencode): preserve Cerebras completion limit (#43736)

Co-authored-by: Aiden <rekram1-node@users.noreply.github.com>
Co-authored-by: ryanl-cerebras <230249923+ryanl-cerebras@users.noreply.github.com>

packages/opencode/src/plugin/cerebras.ts

@@ -0,0 +1,11 @@
+import type { Hooks, PluginInput } from "@opencode-ai/plugin"
+
+export async function CerebrasPlugin(_input: PluginInput): Promise<Hooks> {
+  return {
+    "chat.params": async (input, output) => {
+      if (input.model.api.npm !== "@ai-sdk/cerebras") return
+      if (output.options.max_completion_tokens === undefined) return
+      output.maxOutputTokens = undefined
+    },
+  }
+}

packages/opencode/src/plugin/index.ts

@@ -20,6 +20,7 @@ import { CloudflareAIGatewayAuthPlugin, CloudflareWorkersAuthPlugin } from "./cl
 import { AzureAuthPlugin } from "./azure"
 import { DigitalOceanAuthPlugin } from "./digitalocean"
 import { XaiAuthPlugin } from "./xai"
+import { CerebrasPlugin } from "./cerebras"
 import { SnowflakeCortexAuthPlugin } from "./snowflake-cortex"
 import { Effect, Layer, Context } from "effect"
 import { EffectBridge } from "@/effect/bridge"
@@ -80,6 +81,7 @@ function internalPlugins(flags: RuntimeFlags.Info): PluginInstance[] {
     DigitalOceanAuthPlugin,
     SnowflakeCortexAuthPlugin,
     XaiAuthPlugin,
+    CerebrasPlugin,
   ]
 }
 

packages/opencode/test/plugin/cerebras.test.ts

@@ -0,0 +1,47 @@
+import { describe, expect, test } from "bun:test"
+import type { Hooks, PluginInput } from "@opencode-ai/plugin"
+import { CerebrasPlugin } from "../../src/plugin/cerebras"
+
+type ChatParams = NonNullable<Hooks["chat.params"]>
+
+function input(npm: string) {
+  return {
+    model: { api: { npm } },
+  } as Parameters<ChatParams>[0]
+}
+
+function output(options: Record<string, unknown>) {
+  return {
+    maxOutputTokens: 32_000,
+    options,
+  } as Parameters<ChatParams>[1]
+}
+
+describe("CerebrasPlugin", () => {
+  test("omits the generic output cap when max_completion_tokens is configured", async () => {
+    const hook = (await CerebrasPlugin({} as PluginInput))["chat.params"]!
+    const params = output({ max_completion_tokens: 64 })
+
+    await hook(input("@ai-sdk/cerebras"), params)
+
+    expect(params.maxOutputTokens).toBeUndefined()
+  })
+
+  test("preserves the generic output cap without max_completion_tokens", async () => {
+    const hook = (await CerebrasPlugin({} as PluginInput))["chat.params"]!
+    const params = output({})
+
+    await hook(input("@ai-sdk/cerebras"), params)
+
+    expect(params.maxOutputTokens).toBe(32_000)
+  })
+
+  test("does not change other providers", async () => {
+    const hook = (await CerebrasPlugin({} as PluginInput))["chat.params"]!
+    const params = output({ max_completion_tokens: 64 })
+
+    await hook(input("@ai-sdk/openai"), params)
+
+    expect(params.maxOutputTokens).toBe(32_000)
+  })
+})