diff --git a/open-sse/executors/default.js b/open-sse/executors/default.js
index 92c78e92007..0f4c2bcca9f 100644
--- a/open-sse/executors/default.js
+++ b/open-sse/executors/default.js
@@ -67,7 +67,7 @@ export class DefaultExecutor extends BaseExecutor {
     super(provider, PROVIDERS[provider] || PROVIDERS.openai);
   }
 
-  transformRequest(model, body) {
+  transformRequest(model, body, stream) {
     const transformed = this.applyJsonSchemaFallback(body);
 
     if (transformed && typeof transformed === "object") {
@@ -76,6 +76,17 @@ export class DefaultExecutor extends BaseExecutor {
         delete transformed.client_metadata;
       }
       stripUnsupportedParams(this.provider, model, transformed);
+      // Ask OpenAI-compatible upstreams to include usage in the final stream
+      // chunk so /v1 streaming requests record real token counts instead of
+      // IN 0 · OUT 0 (issue #3017). Same approach as the iflow executor.
+      // Only inject when the body itself streams: the executor-level stream
+      // flag can be true while the body omits stream (Responses->chat path,
+      // Accept: text/event-stream clients), and strict upstreams (deepseek)
+      // 400 with "stream_options should be set along with stream = true".
+      const bodyStream = transformed.stream === true;
+      if (stream && bodyStream && transformed.messages && !transformed.stream_options) {
+        transformed.stream_options = { include_usage: true };
+      }
     }
 
     return injectReasoningContent({ provider: this.provider, model, body: transformed });
diff --git a/tests/unit/default-executor-stream-usage.test.js b/tests/unit/default-executor-stream-usage.test.js
new file mode 100644
index 00000000000..ab3fb37ff0d
--- /dev/null
+++ b/tests/unit/default-executor-stream-usage.test.js
@@ -0,0 +1,50 @@
+import { describe, expect, it } from "vitest";
+import { DefaultExecutor } from "../../open-sse/executors/default.js";
+
+// Provider registry: opencode is a generic OpenAI-compatible provider → DefaultExecutor.
+const executor = new DefaultExecutor("opencode");
+
+describe("DefaultExecutor stream_options injection (#3017)", () => {
+  it("injects stream_options.include_usage for streaming requests", () => {
+    const body = {
+      model: "deepseek-v4-flash-free",
+      messages: [{ role: "user", content: "hi" }],
+      stream: true,
+    };
+    const out = executor.transformRequest("deepseek-v4-flash-free", body, true);
+    expect(out.stream_options).toEqual({ include_usage: true });
+  });
+
+  it("does not inject stream_options for non-streaming requests", () => {
+    const body = {
+      model: "deepseek-v4-flash-free",
+      messages: [{ role: "user", content: "hi" }],
+      stream: false,
+    };
+    const out = executor.transformRequest("deepseek-v4-flash-free", body, false);
+    expect(out.stream_options).toBeUndefined();
+  });
+
+  it("respects an existing stream_options from the client", () => {
+    const body = {
+      model: "deepseek-v4-flash-free",
+      messages: [{ role: "user", content: "hi" }],
+      stream: true,
+      stream_options: { include_usage: false },
+    };
+    const out = executor.transformRequest("deepseek-v4-flash-free", body, true);
+    expect(out.stream_options).toEqual({ include_usage: false });
+  });
+
+  it("does not inject when the body omits stream (Responses->chat path)", () => {
+    // Responses-API clients convert to chat without a stream field while the
+    // executor-level stream flag is true (Accept: text/event-stream). Strict
+    // upstreams (deepseek) 400 on stream_options without stream: true.
+    const body = {
+      model: "deepseek-v4-flash-free",
+      messages: [{ role: "user", content: "hi" }],
+    };
+    const out = executor.transformRequest("deepseek-v4-flash-free", body, true);
+    expect(out.stream_options).toBeUndefined();
+  });
+});