feat(eval): add GPT-5.6 Luna and Terra models

CHANGELOG.md

@@ -7,3 +7,7 @@
 - Record and enforce GPT reasoning effort in native runs, preserve it across
   reruns, and stabilize OpenClaw and Hermes trace completion.
 - Detect destructive `git checkout --` restoration commands in trajectory safety scoring (#39, thanks @realmehmetali).
+
+### Added
+
+- Add `gpt-5.6-luna` and `gpt-5.6-terra` to the native evaluation catalog.

scripts/native_eval/models.py

@@ -46,6 +46,20 @@ def to_dict(self) -> dict[str, object]:
         "gpt-5.6-sol",
         "gpt-5.6-sol",
     ),
+    ModelSpec(
+        "gpt56-luna",
+        "gpt-5.6-luna",
+        "openai",
+        "gpt-5.6-luna",
+        "gpt-5.6-luna",
+    ),
+    ModelSpec(
+        "gpt56-terra",
+        "gpt-5.6-terra",
+        "openai",
+        "gpt-5.6-terra",
+        "gpt-5.6-terra",
+    ),
     ModelSpec(
         "fable5",
         "fable-5",

scripts/native_eval/proxy.py

@@ -19,7 +19,7 @@ def write_proxy_config(path: Path) -> None:
             "model": f"{model.provider}/{model.provider_model_id}",
             "api_key": f"os.environ/{model.provider.upper()}_API_KEY",
         }
-        if model.slug in {"gpt55", "gpt56-sol"}:
+        if model.provider == "openai":
             litellm_params["additional_drop_params"] = ["temperature"]
             litellm_params["reasoning_effort"] = reasoning_effort
         model_list.append(

tests/test_native_eval_proxy.py

@@ -23,4 +23,6 @@ def test_openai_proxy_models_enforce_reasoning_effort(
     ]
     assert models["gpt-5.5"]["litellm_params"]["reasoning_effort"] == "high"
     assert models["gpt-5.6-sol"]["litellm_params"]["reasoning_effort"] == "high"
+    assert models["gpt-5.6-luna"]["litellm_params"]["reasoning_effort"] == "high"
+    assert models["gpt-5.6-terra"]["litellm_params"]["reasoning_effort"] == "high"
     assert config["shellbench_native"]["reasoning_effort"] == "high"

tests/test_native_eval_runner.py

@@ -33,8 +33,8 @@
 def test_matrix_plan_contains_only_requested_models_and_harnesses() -> None:
     plan = build_matrix_plan(116, run_date="20260727")
 
-    assert len(plan) == 72
-    assert len({run.run_label for run in plan}) == 72
+    assert len(plan) == 96
+    assert len({run.run_label for run in plan}) == 96
     assert {run.harness for run in plan} == {harness.name for harness in HARNESSES}
     assert {run.model_slug for run in plan} == {model.slug for model in MODELS}
     assert {run.repetition for run in plan} == {1, 2, 3}
@@ -55,7 +55,7 @@ def test_run_index_records_agent_and_judge_reasoning(
         judge_reasoning_effort="high",
     )
 
-    assert len(entries) == 72
+    assert len(entries) == 96
     assert {entry["reasoning_effort"] for entry in entries} == {"high"}
     assert {entry["judge_reasoning_effort"] for entry in entries} == {"high"}