diff --git a/mteb/benchmarks/_create_table.py b/mteb/benchmarks/_create_table.py
index fba9eae7e1..a517a36d62 100644
--- a/mteb/benchmarks/_create_table.py
+++ b/mteb/benchmarks/_create_table.py
@@ -270,7 +270,6 @@ def _create_summary_table_mean_public_private(
     Returns:
         DataFrame with model summaries, ready for styling in the leaderboard
     """
-    print("all tasks:", benchmark_results.task_names)
     data = benchmark_results.to_dataframe(format="long")
 
     if data.empty:
@@ -279,12 +278,9 @@ def _create_summary_table_mean_public_private(
         )
         return no_results_frame
     public_task_name = benchmark_results.filter_tasks(is_public=True).task_names
-    print("Public tasks:", public_task_name)
     private_task_name = benchmark_results.filter_tasks(is_public=False).task_names
-    print("Private tasks:", private_task_name)
     # Convert to DataFrame and pivot
     per_task = data.pivot(index="model_name", columns="task_name", values="score")
-    print(per_task.columns)
 
     # Remove models with no scores
     to_remove = per_task.isna().all(axis="columns")
@@ -384,3 +380,130 @@ def _create_summary_table_mean_public_private(
     joint_table.insert(0, "Rank (Borda)", joint_table.pop("borda_rank"))
 
     return joint_table
+
+
+def _create_summary_table_mean_subset(
+    benchmark_results: BenchmarkResults,
+) -> pd.DataFrame:
+    """Create summary table from BenchmarkResults.
+
+    Returns a DataFrame with one row per model containing summary statistics
+    and task type averages. Calculates means where each task-language subset
+    is weighted equally.
+
+    Args:
+        benchmark_results: BenchmarkResults object containing model results
+
+    Returns:
+        DataFrame with model summaries, ready for styling in the leaderboard
+    """
+    data = benchmark_results.to_dataframe(format="long")
+
+    if data.empty:
+        no_results_frame = pd.DataFrame(
+            {"No results": ["You can try relaxing your criteria"]}
+        )
+        return no_results_frame
+
+    # Convert to DataFrame and pivot
+    per_task = data.pivot(index="model_name", columns="task_name", values="score")
+
+    # Remove models with no scores
+    to_remove = per_task.isna().all(axis="columns")
+    if to_remove.all():
+        no_results_frame = pd.DataFrame(
+            {"No results": ["You can try relaxing your criteria"]}
+        )
+        return no_results_frame
+
+    models_to_remove = list(per_task[to_remove].index)
+    per_task = per_task.drop(models_to_remove, axis=0)
+
+    # Calculate means by task type
+    mean_per_type = _get_means_per_types(per_task)
+    mean_per_type = mean_per_type.pivot(
+        index="model_name", columns="task_type", values="score"
+    )
+    mean_per_type.columns = [
+        _split_on_capital(column) for column in mean_per_type.columns
+    ]
+
+    # Calculate subset means (each task-language combination weighted equally)
+    detailed_data = benchmark_results.to_dataframe(
+        aggregation_level="subset", format="long"
+    )
+    overall_subset_mean = detailed_data.groupby("model_name")["score"].mean()
+
+    per_subset = detailed_data.pivot(
+        index="model_name", columns=["task_name", "subset"], values="score"
+    )
+
+    # Build joint table
+    joint_table = mean_per_type.copy()
+    joint_table = joint_table.drop(models_to_remove, axis=0)
+    joint_table.insert(0, "mean(subset)", overall_subset_mean)
+    joint_table["borda_rank"] = _get_borda_rank(per_subset)
+    joint_table = joint_table.sort_values("mean(subset)", ascending=False)
+    joint_table = joint_table.reset_index()
+
+    # Add model metadata
+    model_metas = joint_table["model_name"].map(_failsafe_get_model_meta)
+    joint_table = joint_table[model_metas.notna()]
+    joint_table["model_link"] = model_metas.map(lambda m: m.reference)
+
+    # Insert model metadata columns
+    joint_table.insert(
+        1,
+        "Max Tokens",
+        model_metas.map(lambda m: _format_max_tokens(m.max_tokens)),
+    )
+    joint_table.insert(
+        1,
+        "Embedding Dimensions",
+        model_metas.map(lambda m: str(int(m.embed_dim)) if m.embed_dim else "Unknown"),
+    )
+    joint_table.insert(
+        1,
+        "Number of Parameters",
+        model_metas.map(lambda m: _format_n_parameters(m.n_parameters)),
+    )
+    joint_table.insert(
+        1,
+        "Memory Usage (MB)",
+        model_metas.map(
+            lambda m: str(int(m.memory_usage_mb)) if m.memory_usage_mb else "Unknown"
+        ),
+    )
+
+    # Add zero-shot percentage
+    tasks = get_tasks(tasks=list(data["task_name"].unique()))
+    joint_table.insert(
+        1, "Zero-shot", model_metas.map(lambda m: m.zero_shot_percentage(tasks))
+    )
+    joint_table["Zero-shot"] = joint_table["Zero-shot"].fillna(-1)
+
+    # Clean up model names (remove HF organization)
+    joint_table["model_name"] = joint_table["model_name"].map(
+        lambda name: name.split("/")[-1]
+    )
+
+    # Add markdown links to model names
+    name_w_link = (
+        "[" + joint_table["model_name"] + "](" + joint_table["model_link"] + ")"
+    )
+    joint_table["model_name"] = joint_table["model_name"].mask(
+        joint_table["model_link"].notna(), name_w_link
+    )
+    joint_table = joint_table.drop(columns=["model_link"])
+
+    # Rename columns
+    rename_dict = {
+        "model_name": "Model",
+        "mean(subset)": "Mean (Subset)",
+    }
+    joint_table = joint_table.rename(columns=rename_dict)
+
+    # Move borda rank to front
+    joint_table.insert(0, "Rank (Borda)", joint_table.pop("borda_rank"))
+
+    return joint_table
diff --git a/mteb/benchmarks/benchmark.py b/mteb/benchmarks/benchmark.py
index c6a570be8b..371096aa86 100644
--- a/mteb/benchmarks/benchmark.py
+++ b/mteb/benchmarks/benchmark.py
@@ -11,6 +11,7 @@
     _create_per_task_table_from_benchmark_results,
     _create_summary_table_from_benchmark_results,
     _create_summary_table_mean_public_private,
+    _create_summary_table_mean_subset,
 )
 from mteb.load_results.load_results import load_results
 
@@ -98,3 +99,11 @@ def _create_summary_table(
     ) -> pd.DataFrame:
         """Create summary table. Called by the leaderboard app."""
         return _create_summary_table_mean_public_private(benchmark_results)
+
+
+class HUMEBenchmark(Benchmark):
+    def _create_summary_table(
+        self, benchmark_results: BenchmarkResults
+    ) -> pd.DataFrame:
+        """Create summary table. Called by the leaderboard app."""
+        return _create_summary_table_mean_subset(benchmark_results)
diff --git a/mteb/benchmarks/benchmarks/benchmarks.py b/mteb/benchmarks/benchmarks/benchmarks.py
index 882b8d15c7..3feb4b0cb3 100644
--- a/mteb/benchmarks/benchmarks/benchmarks.py
+++ b/mteb/benchmarks/benchmarks/benchmarks.py
@@ -4,7 +4,7 @@
 
 from pydantic import AnyUrl, BeforeValidator, TypeAdapter
 
-from mteb.benchmarks.benchmark import Benchmark
+from mteb.benchmarks.benchmark import Benchmark, HUMEBenchmark
 from mteb.overview import MTEBTasks, get_task, get_tasks
 
 if TYPE_CHECKING:
@@ -2303,7 +2303,7 @@
 )
 
 
-HUME = Benchmark(
+HUME = HUMEBenchmark(
     name="HUME(v1)",
     display_name="Human Benchmark",
     # icon="https://raw.githubusercontent.com/huggingface/benchmarks/main/benchmarks/assets/hume.png",
@@ -2326,7 +2326,13 @@
             "HUMESTSBenchmark",
             "HUMESTS22",
         ],
-        languages=["eng-Latn", "ara-Arab", "rus-Cyrl", "dan-Latn", "nob-Latn"],
+        languages=[
+            "eng-Latn",
+            "ara-Arab",
+            "rus-Cyrl",
+            "dan-Latn",
+            "nob-Latn",
+        ],
     ),
     description="The HUME benchmark is designed to evaluate the performance of text embedding models and humans on a comparable set of tasks. This captures areas where models perform better than human annotators and the reverse. In the paper, we go further into the analysis and what conclusions can be drawn.",
     reference="Coming soon (in review)",
diff --git a/mteb/leaderboard/benchmark_selector.py b/mteb/leaderboard/benchmark_selector.py
index 148dcbb114..0e5f07bae1 100644
--- a/mteb/leaderboard/benchmark_selector.py
+++ b/mteb/leaderboard/benchmark_selector.py
@@ -34,7 +34,9 @@ class MenuEntry:
         name="General Purpose",
         description="",
         open=False,
-        benchmarks=mteb.get_benchmarks(["MTEB(Multilingual, v2)", "MTEB(eng, v2)"])
+        benchmarks=mteb.get_benchmarks(
+            ["MTEB(Multilingual, v2)", "MTEB(eng, v2)", "HUME(v1)"]
+        )
         + [
             MenuEntry(
                 "Image",
diff --git a/mteb/tasks/Classification/multilingual/HUMEMultilingualSentimentClassification.py b/mteb/tasks/Classification/multilingual/HUMEMultilingualSentimentClassification.py
index a021b120a1..5c082f7fe3 100644
--- a/mteb/tasks/Classification/multilingual/HUMEMultilingualSentimentClassification.py
+++ b/mteb/tasks/Classification/multilingual/HUMEMultilingualSentimentClassification.py
@@ -7,7 +7,7 @@
 _LANGUAGES = {
     "eng": ["eng-Latn"],
     "ara": ["ara-Arab"],
-    "nor": ["nor-Latn"],
+    "nor": ["nob-Latn"],
     "rus": ["rus-Cyrl"],
 }