diff --git a/packages/cli/src/commands/benchmark.test.ts b/packages/cli/src/commands/benchmark.test.ts
index 8e9a017fe8..793b9c617f 100644
--- a/packages/cli/src/commands/benchmark.test.ts
+++ b/packages/cli/src/commands/benchmark.test.ts
@@ -1,12 +1,15 @@
 import { runCommand } from "citty";
-import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { existsSync, mkdirSync, mkdtempSync, rmSync, writeFileSync } from "node:fs";
 import { tmpdir } from "node:os";
-import { join } from "node:path";
+import { dirname, join } from "node:path";
 import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 import { consumeCommandResult } from "../utils/commandResult.js";
 import benchmark from "./benchmark.js";
 
-const producer = vi.hoisted(() => ({ executeRenderJob: vi.fn<() => Promise<void>>() }));
+const producer = vi.hoisted(() => ({
+  executeRenderJob:
+    vi.fn<(_job: unknown, _projectDir: string, outputPath: string) => Promise<void>>(),
+}));
 vi.mock("../utils/producer.js", () => ({
   loadProducer: async () => ({
     createRenderJob: (config: unknown) => ({ config }),
@@ -90,4 +93,22 @@ describe.each([true, false])("benchmark (json=%s)", (json) => {
 
     expect(await runBenchmark(json)).toMatchObject({ exitCode: 0, kind: "success" });
   });
+
+  it("isolates output between invocations and removes temporary renders", async () => {
+    producer.executeRenderJob.mockImplementation(async (_job, _projectDir, outputPath) => {
+      mkdirSync(dirname(outputPath), { recursive: true });
+      writeFileSync(outputPath, "benchmark output");
+    });
+
+    await runBenchmark(json);
+    await runBenchmark(json);
+
+    const outputDirs = [0, 5].map((index) => {
+      const outputPath = producer.executeRenderJob.mock.calls[index]?.[2];
+      expect(outputPath).toBeDefined();
+      return dirname(outputPath ?? "");
+    });
+    expect(outputDirs[0]).not.toBe(outputDirs[1]);
+    expect(outputDirs.every((dir) => !existsSync(dir))).toBe(true);
+  });
 });
diff --git a/packages/cli/src/commands/benchmark.ts b/packages/cli/src/commands/benchmark.ts
index 7e42722123..79014df9d2 100644
--- a/packages/cli/src/commands/benchmark.ts
+++ b/packages/cli/src/commands/benchmark.ts
@@ -1,14 +1,15 @@
 import { failCommand, setCommandExitCode } from "../utils/commandResult.js";
 import { defineCommand } from "citty";
 import type { Example } from "./_examples.js";
-import { existsSync, statSync } from "node:fs";
+import { existsSync, mkdtempSync, rmSync, statSync } from "node:fs";
+import { tmpdir } from "node:os";
 
 export const examples: Example[] = [
   ["Run benchmarks with default settings (3 runs)", "hyperframes benchmark"],
   ["Run 5 iterations per config", "hyperframes benchmark --runs 5"],
   ["Output results as JSON", "hyperframes benchmark --json"],
 ];
-import { resolve, join } from "node:path";
+import { join } from "node:path";
 import { resolveProject } from "../utils/project.js";
 import { loadProducer } from "../utils/producer.js";
 import { c } from "../ui/colors.js";
@@ -71,7 +72,7 @@ export default defineCommand({
     const jsonOutput = args.json ?? false;
 
     // ── Temp output for benchmark renders ────────────────────────────────
-    const benchDir = resolve("renders", ".benchmark");
+    const benchDir = mkdtempSync(join(tmpdir(), "hyperframes-benchmark-"));
 
     // ── Load producer ────────────────────────────────────────────────────
     let producer: Awaited<ReturnType<typeof loadProducer>> | null = null;
@@ -159,6 +160,8 @@ export default defineCommand({
       results.push({ config, runs, failures, avgTime, avgSize });
     }
 
+    rmSync(benchDir, { recursive: true, force: true });
+
     if (results.every((result) => result.runs.length === 0)) {
       setCommandExitCode(1);
     }