diff --git a/bench/perf-table.md b/bench/perf-table.md
new file mode 100644
index 00000000..4c33f9fc
--- /dev/null
+++ b/bench/perf-table.md
@@ -0,0 +1,13 @@
+<!-- generated by scripts/gen-perf-table.py from bench/results.json (2026-05-21T21:52:31Z) -->
+
+## Performance
+
+Wall-clock time (lower is better). Baseline = Node.
+
+| Benchmark | ilo-jit | ilo-vm | Node | Python |
+| --- | --- | --- | --- | --- |
+| fib | 25.4 ms (×4.7) | 70.7 ms (×13.2) | 5.4 ms (×1) | 56.9 ms (×10.6) |
+| hof | 455.4 ms (×6.5) | 160.7 ms (×2.3) | 70.3 ms (×1) | 125.6 ms (×1.8) |
+| listproc | 1.5 ms (×1) | 65.7 ms (×46.4) | 1.4 ms (×1) | 118.8 ms (×84) |
+| pattern-match | 3.1 ms (×1.2) | 146.3 ms (×55.8) | 2.6 ms (×1) | 151.5 ms (×57.8) |
+| sum-loop | 1.4 ms (×1.4) | 51.5 ms (×52.6) | 978 µs (×1) | 63.0 ms (×64.4) |
diff --git a/scripts/gen-perf-table.py b/scripts/gen-perf-table.py
new file mode 100755
index 00000000..2ba21116
--- /dev/null
+++ b/scripts/gen-perf-table.py
@@ -0,0 +1,101 @@
+#!/usr/bin/env python3
+"""Generate bench/perf-table.md from bench/results.json.
+
+Usage:
+    python3 scripts/gen-perf-table.py
+
+Reads  bench/results.json  relative to the repo root and writes
+bench/perf-table.md.  Run this script whenever bench/results.json is
+updated (e.g. in CI after bench/run.sh).
+
+Website integration
+-------------------
+If a static-site generator is added under site/, include the generated
+file with a snippet such as:
+
+    --8<-- "bench/perf-table.md"       (mkdocs-material)
+    {% include "../../bench/perf-table.md" %}   (Jinja / Jekyll)
+
+The file contains only the table and a one-line header comment so it
+can be embedded directly without further processing.
+"""
+
+import json
+import pathlib
+import sys
+
+REPO_ROOT = pathlib.Path(__file__).parent.parent
+RESULTS = REPO_ROOT / "bench" / "results.json"
+OUT = REPO_ROOT / "bench" / "perf-table.md"
+
+
+def fmt_us(value_us: int) -> str:
+    """Format microseconds: show ms when >= 1000, otherwise us."""
+    if value_us >= 1000:
+        ms = value_us / 1000
+        return f"{ms:.1f} ms"
+    return f"{value_us} µs"
+
+
+def ratio(value: int, baseline: int) -> str:
+    """Return a ×N.N relative-to-baseline string."""
+    r = value / baseline
+    if abs(r - round(r)) < 0.05:
+        return f"×{r:.0f}"
+    return f"×{r:.1f}"
+
+
+def main() -> None:
+    if not RESULTS.exists():
+        print(f"ERROR: {RESULTS} not found", file=sys.stderr)
+        sys.exit(1)
+
+    data = json.loads(RESULTS.read_text())
+    generated = data.get("generated", "unknown")
+    benchmarks = data["benchmarks"]
+
+    # Determine column order: ilo-jit first, then others alphabetically
+    all_impls: list[str] = []
+    for times in benchmarks.values():
+        for impl in times:
+            if impl not in all_impls:
+                all_impls.append(impl)
+
+    preferred = ["ilo-jit", "ilo-vm"]
+    ordered = [i for i in preferred if i in all_impls] + sorted(
+        i for i in all_impls if i not in preferred
+    )
+
+    lines: list[str] = [
+        f"<!-- generated by scripts/gen-perf-table.py from bench/results.json ({generated}) -->",
+        "",
+        "## Performance",
+        "",
+        "Wall-clock time (lower is better). Baseline = Node.",
+        "",
+    ]
+
+    # Header row
+    header = "| Benchmark | " + " | ".join(ordered) + " |"
+    separator = "| --- | " + " | ".join(["---"] * len(ordered)) + " |"
+    lines += [header, separator]
+
+    for bench, times in sorted(benchmarks.items()):
+        baseline = times.get("Node") or next(iter(times.values()))
+        cells: list[str] = []
+        for impl in ordered:
+            if impl not in times:
+                cells.append("—")
+            else:
+                t = times[impl]
+                r = ratio(t, baseline)
+                cells.append(f"{fmt_us(t)} ({r})")
+        lines.append("| " + bench + " | " + " | ".join(cells) + " |")
+
+    lines += [""]
+    OUT.write_text("\n".join(lines))
+    print(f"Wrote {OUT}")
+
+
+if __name__ == "__main__":
+    main()