diff --git a/targets/sv-comp/scripts/README.md b/targets/sv-comp/scripts/README.md
index 22ce37c..1d7855b 100644
--- a/targets/sv-comp/scripts/README.md
+++ b/targets/sv-comp/scripts/README.md
@@ -98,6 +98,10 @@ scripts/
 ./svcomp analyze compare runs/run_A/results/results_valid-assert.prp_*.json \
                          runs/run_B/results/results_valid-assert.prp_*.json
 
+# Compare two BenchExec XML files or run directories
+./svcomp analyze benchdiff results/old-run results/new-run
+./svcomp analyze benchdiff results/old-run results/new-run --format markdown
+
 # Show test case statistics
 ./svcomp analyze stats
 ```
diff --git a/targets/sv-comp/scripts/commands/analyze.py b/targets/sv-comp/scripts/commands/analyze.py
index ce71578..8147257 100644
--- a/targets/sv-comp/scripts/commands/analyze.py
+++ b/targets/sv-comp/scripts/commands/analyze.py
@@ -65,6 +65,49 @@ def compare(ctx, current_file, reference_file):
         ctx.exit(1)
 
 
+@analyze.command(name='benchdiff')
+@click.argument('reference_path', type=click.Path(exists=True))
+@click.argument('current_path', type=click.Path(exists=True))
+@click.option(
+    '--format',
+    'output_format',
+    type=click.Choice(['text', 'markdown', 'json']),
+    default='text',
+    help='Output format',
+)
+@click.option(
+    '--category-only',
+    is_flag=True,
+    help='Ignore status changes if the BenchExec category stayed the same',
+)
+@click.pass_context
+def benchdiff(ctx, reference_path, current_path, output_format, category_only):
+    """Compare BenchExec XML result files or run directories.
+
+    Shows changed per-task outcomes between REFERENCE_PATH and CURRENT_PATH.
+    Each path may point to one .xml/.xml.bz2 file or to a directory containing
+    BenchExec result XML files.
+    """
+    from lib import compare_benchexec_results, diff_to_json, format_benchexec_diff
+
+    try:
+        diff = compare_benchexec_results(
+            [reference_path],
+            [current_path],
+            include_status_changes=not category_only,
+        )
+        if output_format == 'json':
+            click.echo(diff_to_json(diff))
+        else:
+            click.echo(format_benchexec_diff(diff, markdown=output_format == 'markdown'))
+    except click.exceptions.Exit:
+        raise
+    except Exception as e:
+        click.echo(f"Error comparing BenchExec results: {e}", err=True)
+        logger.exception("BenchExec comparison failed")
+        ctx.exit(1)
+
+
 @analyze.command(name='stats')
 @click.option('--benchmark-dir', type=click.Path(exists=True), help='Path to sv-benchmarks directory')
 @click.pass_context
diff --git a/targets/sv-comp/scripts/lib/__init__.py b/targets/sv-comp/scripts/lib/__init__.py
index 818811d..9598349 100644
--- a/targets/sv-comp/scripts/lib/__init__.py
+++ b/targets/sv-comp/scripts/lib/__init__.py
@@ -20,6 +20,13 @@
 from .witness import generate_and_validate_witness
 from .analysis.context_loss import main as analyze_context_loss
 from .comparison import compare_results
+from .benchexec_comparison import (
+    collect_result_files,
+    load_benchexec_results,
+    compare_benchexec_results,
+    format_benchexec_diff,
+    diff_to_json,
+)
 from .utils import ci_print
 from .cfg_metrics import (
     check_cfg_extractor_available,
@@ -55,6 +62,11 @@
     "analyze_context_loss",
     # Comparison
     "compare_results",
+    "collect_result_files",
+    "load_benchexec_results",
+    "compare_benchexec_results",
+    "format_benchexec_diff",
+    "diff_to_json",
     # Utils
     "ci_print",
     # CFG Metrics
diff --git a/targets/sv-comp/scripts/lib/benchexec_comparison.py b/targets/sv-comp/scripts/lib/benchexec_comparison.py
new file mode 100644
index 0000000..b034913
--- /dev/null
+++ b/targets/sv-comp/scripts/lib/benchexec_comparison.py
@@ -0,0 +1,232 @@
+"""Compare BenchExec result XML files."""
+
+from __future__ import annotations
+
+import bz2
+import json
+import xml.etree.ElementTree as ET
+from collections import Counter
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Iterable
+
+
+@dataclass(frozen=True, order=True)
+class BenchmarkKey:
+    """Stable identity of one benchmark task in one property."""
+
+    task: str
+    property: str
+    expected_verdict: str
+
+
+@dataclass(frozen=True)
+class BenchmarkOutcome:
+    """Outcome reported by BenchExec for one task."""
+
+    category: str
+    status: str
+    cputime: str
+    walltime: str
+    source: str
+    display_name: str
+    runset: str
+
+    @property
+    def label(self) -> str:
+        if self.category and self.status:
+            return f"{self.category}/{self.status}"
+        return self.category or self.status or "missing"
+
+
+def collect_result_files(paths: Iterable[str | Path]) -> list[Path]:
+    """Collect BenchExec XML result files from files or directories."""
+
+    result_files: list[Path] = []
+    for raw_path in paths:
+        path = Path(raw_path)
+        if path.is_dir():
+            result_files.extend(path.rglob("*.xml"))
+            result_files.extend(path.rglob("*.xml.bz2"))
+        elif path.is_file() and (path.name.endswith(".xml") or path.name.endswith(".xml.bz2")):
+            result_files.append(path)
+        else:
+            raise FileNotFoundError(f"No BenchExec XML result found at {path}")
+
+    unique_files = sorted(set(result_files))
+    if not unique_files:
+        raise FileNotFoundError("No BenchExec XML result files found")
+    return unique_files
+
+
+def load_benchexec_results(paths: Iterable[str | Path]) -> dict[BenchmarkKey, BenchmarkOutcome]:
+    """Load task outcomes from BenchExec result XML files."""
+
+    results: dict[BenchmarkKey, BenchmarkOutcome] = {}
+    for result_file in collect_result_files(paths):
+        root = _parse_xml(result_file)
+        display_name = root.attrib.get("displayName", "")
+        runset = root.attrib.get("name", "")
+
+        for run in root.findall("run"):
+            columns = {column.attrib.get("title", ""): column.attrib.get("value", "") for column in run.findall("column")}
+            key = BenchmarkKey(
+                task=_normalize_task_name(run.attrib.get("name", "")),
+                property=_extract_property(run),
+                expected_verdict=run.attrib.get("expectedVerdict", ""),
+            )
+            if key in results:
+                raise ValueError(f"Duplicate benchmark result for {key}")
+
+            results[key] = BenchmarkOutcome(
+                category=columns.get("category", ""),
+                status=columns.get("status", ""),
+                cputime=columns.get("cputime", ""),
+                walltime=columns.get("walltime", ""),
+                source=str(result_file),
+                display_name=display_name,
+                runset=runset,
+            )
+
+    return results
+
+
+def compare_benchexec_results(
+    reference_paths: Iterable[str | Path],
+    current_paths: Iterable[str | Path],
+    *,
+    include_status_changes: bool = True,
+) -> dict:
+    """Compare two BenchExec result sets and return a serializable diff."""
+
+    reference = load_benchexec_results(reference_paths)
+    current = load_benchexec_results(current_paths)
+
+    changes = []
+    for key in sorted(set(reference) | set(current)):
+        old = reference.get(key)
+        new = current.get(key)
+
+        if old is None:
+            transition_changed = True
+        elif new is None:
+            transition_changed = True
+        elif include_status_changes:
+            transition_changed = (old.category, old.status) != (new.category, new.status)
+        else:
+            transition_changed = old.category != new.category
+
+        if transition_changed:
+            changes.append(_change_to_dict(key, old, new))
+
+    return {
+        "reference": _summarize(reference),
+        "current": _summarize(current),
+        "changes": changes,
+        "transition_counts": dict(Counter(change["transition"] for change in changes)),
+    }
+
+
+def format_benchexec_diff(diff: dict, *, markdown: bool = False) -> str:
+    """Format a BenchExec comparison for humans."""
+
+    reference = diff["reference"]
+    current = diff["current"]
+    changes = diff["changes"]
+    gained = sum(1 for change in changes if change["from_category"] != "correct" and change["to_category"] == "correct")
+    lost = sum(1 for change in changes if change["from_category"] == "correct" and change["to_category"] != "correct")
+
+    if markdown:
+        lines = [
+            f"- Reference: {reference['correct']}/{reference['total']} correct",
+            f"- Current: {current['correct']}/{current['total']} correct",
+            f"- Outcome changes: {len(changes)} ({gained} gained correct, {lost} lost correct)",
+            "",
+        ]
+        for change in changes:
+            lines.append(
+                f"- `{change['property']}` `{change['task']}`: "
+                f"`{change['from']}` -> `{change['to']}`"
+            )
+        return "\n".join(lines)
+
+    lines = [
+        f"Reference: {reference['correct']}/{reference['total']} correct",
+        f"Current:   {current['correct']}/{current['total']} correct",
+        f"Changes:   {len(changes)} ({gained} gained correct, {lost} lost correct)",
+        "",
+    ]
+    for change in changes:
+        lines.append(
+            f"{change['property']:22} {change['task']:80} "
+            f"{change['from']} -> {change['to']}"
+        )
+    return "\n".join(lines)
+
+
+def diff_to_json(diff: dict) -> str:
+    """Serialize a BenchExec comparison as stable JSON."""
+
+    return json.dumps(diff, indent=2, sort_keys=True)
+
+
+def _parse_xml(path: Path) -> ET.Element:
+    if path.name.endswith(".bz2"):
+        with bz2.open(path, "rb") as xml_file:
+            return ET.fromstring(xml_file.read())
+
+    return ET.parse(path).getroot()
+
+
+def _normalize_task_name(raw_name: str) -> str:
+    marker = "sv-benchmarks/java/"
+    if marker in raw_name:
+        return raw_name.split(marker, 1)[1]
+    return raw_name
+
+
+def _extract_property(run: ET.Element) -> str:
+    prop = run.attrib.get("properties", "")
+    if prop:
+        return prop
+
+    property_file = run.attrib.get("propertyFile", "")
+    if property_file:
+        return Path(property_file).stem
+
+    return ""
+
+
+def _summarize(results: dict[BenchmarkKey, BenchmarkOutcome]) -> dict:
+    categories = Counter(outcome.category or "missing" for outcome in results.values())
+    return {
+        "total": len(results),
+        "correct": categories.get("correct", 0),
+        "categories": dict(sorted(categories.items())),
+        "display_names": sorted({outcome.display_name for outcome in results.values() if outcome.display_name}),
+        "runsets": sorted({outcome.runset for outcome in results.values() if outcome.runset}),
+    }
+
+
+def _change_to_dict(
+    key: BenchmarkKey,
+    old: BenchmarkOutcome | None,
+    new: BenchmarkOutcome | None,
+) -> dict:
+    from_category = old.category if old else "missing"
+    to_category = new.category if new else "missing"
+    from_label = old.label if old else "missing"
+    to_label = new.label if new else "missing"
+
+    return {
+        "task": key.task,
+        "property": key.property,
+        "expected_verdict": key.expected_verdict,
+        "from": from_label,
+        "to": to_label,
+        "from_category": from_category,
+        "to_category": to_category,
+        "transition": f"{from_category}->{to_category}",
+        "reference_source": old.source if old else "",
+        "current_source": new.source if new else "",
+    }