Add gpt5 benchmark tool preset support

benchmarks/hybridgym_depsearch/run_infer.py

@@ -265,6 +265,10 @@ def _get_tools(self, preset: ToolPresetType = "default") -> list[Tool]:
             from openhands.tools.preset.gemini import get_gemini_tools
 
             return get_gemini_tools(enable_browser=False)
+        elif preset == "gpt5":
+            from openhands.tools.preset.gpt5 import get_gpt5_tools
+
+            return get_gpt5_tools(enable_browser=False)
         elif preset == "planning":
             from openhands.tools.preset.planning import get_planning_tools
 

benchmarks/hybridgym_funcgen/run_infer.py

@@ -366,6 +366,10 @@ def _get_tools(self, preset: ToolPresetType = "default") -> list[Tool]:
             from openhands.tools.preset.gemini import get_gemini_tools
 
             return get_gemini_tools(enable_browser=False)
+        elif preset == "gpt5":
+            from openhands.tools.preset.gpt5 import get_gpt5_tools
+
+            return get_gpt5_tools(enable_browser=False)
         elif preset == "planning":
             from openhands.tools.preset.planning import get_planning_tools
 

benchmarks/hybridgym_funclocalize/run_infer.py

@@ -351,6 +351,10 @@ def _get_tools(self, preset: ToolPresetType = "default") -> list[Tool]:
             from openhands.tools.preset.gemini import get_gemini_tools
 
             return get_gemini_tools(enable_browser=False)
+        elif preset == "gpt5":
+            from openhands.tools.preset.gpt5 import get_gpt5_tools
+
+            return get_gpt5_tools(enable_browser=False)
         elif preset == "planning":
             from openhands.tools.preset.planning import get_planning_tools
 

benchmarks/hybridgym_issuelocalize/run_infer.py

@@ -260,6 +260,10 @@ def _get_tools(self, preset: ToolPresetType = "default") -> list[Tool]:
             from openhands.tools.preset.gemini import get_gemini_tools
 
             return get_gemini_tools(enable_browser=False)
+        elif preset == "gpt5":
+            from openhands.tools.preset.gpt5 import get_gpt5_tools
+
+            return get_gpt5_tools(enable_browser=False)
         elif preset == "planning":
             from openhands.tools.preset.planning import get_planning_tools
 

benchmarks/swebench/run_infer.py

@@ -61,7 +61,7 @@ def get_tools_for_preset(
     """Get the list of tools for the given preset.
 
     Args:
-        preset: The tool preset to use (default, gemini, or planning).
+        preset: The tool preset to use (default, gemini, gpt5, or planning).
         enable_browser: Whether to include browser tools.
 
     Returns:
@@ -71,6 +71,10 @@ def get_tools_for_preset(
         from openhands.tools.preset.gemini import get_gemini_tools
 
         return get_gemini_tools(enable_browser=enable_browser)
+    elif preset == "gpt5":
+        from openhands.tools.preset.gpt5 import get_gpt5_tools
+
+        return get_gpt5_tools(enable_browser=enable_browser)
     elif preset == "planning":
         from openhands.tools.preset.planning import get_planning_tools
 

benchmarks/swebenchmultilingual/run_infer.py

@@ -51,7 +51,7 @@ def get_tools_for_preset(
     """Get the list of tools for the given preset.
 
     Args:
-        preset: The tool preset to use (default, gemini, or planning).
+        preset: The tool preset to use (default, gemini, gpt5, or planning).
         enable_browser: Whether to include browser tools.
 
     Returns:
@@ -61,6 +61,10 @@ def get_tools_for_preset(
         from openhands.tools.preset.gemini import get_gemini_tools
 
         return get_gemini_tools(enable_browser=enable_browser)
+    elif preset == "gpt5":
+        from openhands.tools.preset.gpt5 import get_gpt5_tools
+
+        return get_gpt5_tools(enable_browser=enable_browser)
     elif preset == "planning":
         from openhands.tools.preset.planning import get_planning_tools
 

benchmarks/utils/args_parser.py

@@ -89,10 +89,11 @@ def get_parser(add_llm_config: bool = True) -> argparse.ArgumentParser:
         "--tool-preset",
         type=str,
         default="default",
-        choices=["default", "gemini", "planning"],
+        choices=["default", "gemini", "gpt5", "planning"],
         help=(
             "Tool preset for file editing. 'default' uses FileEditorTool, "
-            "'gemini' uses read_file/write_file/edit/list_directory (default: default)"
+            "'gemini' uses read_file/write_file/edit/list_directory, "
+            "'gpt5' uses apply_patch tool (default: default)"
         ),
     )
     parser.add_argument(

benchmarks/utils/models.py

@@ -14,7 +14,7 @@
 
 
 # Tool preset type for selecting which file editing toolset to use
-ToolPresetType = Literal["default", "gemini", "planning"]
+ToolPresetType = Literal["default", "gemini", "gpt5", "planning"]
 
 
 class EvalMetadata(BaseModel):
@@ -101,6 +101,7 @@ class EvalMetadata(BaseModel):
         description=(
             "Tool preset for file editing. 'default' uses FileEditorTool, "
             "'gemini' uses read_file/write_file/edit/list_directory, "
+            "'gpt5' uses apply_patch tool, "
             "'planning' uses planning-mode tools."
         ),
     )

tests/test_multimodal_phased_build.py

@@ -278,6 +278,7 @@ def test_force_build_forwarded(self, _collect, mock_builder, _bases, mock_assemb
 class TestMultimodalParser:
     def test_defaults(self):
         from benchmarks.swebenchmultimodal.build_images import get_parser
+        from benchmarks.swebenchmultimodal.config import BUILD_DEFAULTS
 
         parser = get_parser()
         args = parser.parse_args([])
@@ -287,4 +288,4 @@ def test_defaults(self):
         assert args.push is False
         assert args.force_build is False
         assert args.n_limit == 0
-        assert args.select is None
+        assert args.select == BUILD_DEFAULTS["select"]

tests/test_tool_presets.py

@@ -0,0 +1,50 @@
+import pytest
+
+from benchmarks.hybridgym_depsearch.run_infer import DepSearchEvaluation
+from benchmarks.hybridgym_funcgen.run_infer import FuncGenEvaluation
+from benchmarks.hybridgym_funclocalize.run_infer import FuncLocalizeEvaluation
+from benchmarks.hybridgym_issuelocalize.run_infer import IssueLocalizeEvaluation
+from benchmarks.swebench.run_infer import get_tools_for_preset as get_swebench_tools
+from benchmarks.swebenchmultilingual.run_infer import (
+    get_tools_for_preset as get_swebenchmultilingual_tools,
+)
+from benchmarks.utils.args_parser import get_parser
+from openhands.tools.preset.gpt5 import get_gpt5_tools
+
+
+@pytest.mark.parametrize(
+    "getter",
+    [
+        get_swebench_tools,
+        get_swebenchmultilingual_tools,
+    ],
+)
+def test_swebench_tool_helpers_support_gpt5(getter):
+    expected = [tool.name for tool in get_gpt5_tools(enable_browser=False)]
+
+    assert [tool.name for tool in getter("gpt5", enable_browser=False)] == expected
+
+
+@pytest.mark.parametrize(
+    "evaluation_cls",
+    [
+        DepSearchEvaluation,
+        FuncGenEvaluation,
+        FuncLocalizeEvaluation,
+        IssueLocalizeEvaluation,
+    ],
+)
+def test_hybridgym_tool_helpers_support_gpt5(evaluation_cls):
+    expected = [tool.name for tool in get_gpt5_tools(enable_browser=False)]
+
+    assert [
+        tool.name for tool in evaluation_cls._get_tools(None, preset="gpt5")
+    ] == expected
+
+
+def test_common_parser_accepts_gpt5_tool_preset():
+    parser = get_parser(add_llm_config=False)
+
+    args = parser.parse_args(["--tool-preset", "gpt5"])
+
+    assert args.tool_preset == "gpt5"