diff --git a/src/routes/chat.ts b/src/routes/chat.ts
index 8896fd7..60b1931 100644
--- a/src/routes/chat.ts
+++ b/src/routes/chat.ts
@@ -8,9 +8,17 @@ import { type OpencodeRuntimeAdapter } from "../runtime/opencode.js";
 import { type ChatSessionBridge } from "../chat/sessionBridge.js";
 import { writeLlmRequestAuditLog } from "../audit/llmRequestAudit.js";
 
+const supportedModels = [
+  "deepseek/deepseek-v4-flash",
+  "deepseek/deepseek-v4-pro",
+] as const;
+
+type SupportedModel = (typeof supportedModels)[number];
+
 const payloadSchema = z.object({
   message: z.string().min(1).max(10000),
   session_id: z.string().max(128).optional(),
+  model: z.enum(supportedModels).optional(),
 });
 
 const abortPayloadSchema = z.object({
@@ -169,6 +177,7 @@ export const buildChatRouter = (
           clientSessionId: requestContext.clientSessionId,
           sessionId: binding.sessionId,
           created,
+          model: parsed.data.model,
           traceId: requestContext.traceId,
           projectId: requestContext.projectId,
         },
@@ -207,6 +216,7 @@ export const buildChatRouter = (
           opencodeSessionId: binding.sessionId,
           clientSessionId,
           message: preparedMessage,
+          model: parsed.data.model,
           traceId: requestContext.traceId,
           projectId: requestContext.projectId,
           signal: abortController.signal,
@@ -329,11 +339,26 @@ const extractSkillAuditInfo = (event: OpencodeEvent) => {
 const hasToolParams = (params: Record<string, unknown>) =>
   Object.keys(params).length > 0;
 
+const toRuntimeModel = (model?: SupportedModel) => {
+  if (!model) {
+    return undefined;
+  }
+  const [providerID, modelID] = model.split("/");
+  if (!providerID || !modelID) {
+    return undefined;
+  }
+  return {
+    providerID,
+    modelID,
+  };
+};
+
 type StreamPromptOptions = {
   runtime: OpencodeRuntimeAdapter;
   opencodeSessionId: string;
   clientSessionId: string;
   message: string;
+  model?: SupportedModel;
   traceId?: string;
   projectId?: string;
   signal?: AbortSignal;
@@ -355,6 +380,7 @@ const streamPromptResponse = async ({
   opencodeSessionId,
   clientSessionId,
   message,
+  model,
   traceId,
   projectId,
   signal,
@@ -437,7 +463,7 @@ const streamPromptResponse = async ({
   });
 
   const promptPromise = runtime
-    .prompt(opencodeSessionId, message)
+    .prompt(opencodeSessionId, message, toRuntimeModel(model))
     .then(() => {
       promptSettled = true;
     })
diff --git a/src/runtime/opencode.ts b/src/runtime/opencode.ts
index f83bdd6..b4b17b3 100644
--- a/src/runtime/opencode.ts
+++ b/src/runtime/opencode.ts
@@ -12,6 +12,11 @@ export type RuntimeHealth = {
   version: string;
 };
 
+type RuntimeModelOverride = {
+  providerID: string;
+  modelID: string;
+};
+
 export class OpencodeRuntimeAdapter {
   private clientPromise: Promise<OpencodeClient> | null = null;
   private closeServer: (() => void) | null = null;
@@ -52,10 +57,11 @@ export class OpencodeRuntimeAdapter {
     return this.messages(sessionId);
   }
 
-  async prompt(sessionId: string, text: string) {
+  async prompt(sessionId: string, text: string, model?: RuntimeModelOverride) {
     const client = await this.ensureClient();
     await client.session.prompt({
       sessionID: sessionId,
+      model,
       parts: [{ type: "text", text }],
     });
   }