diff --git a/services/bench/common/bench_common/model_catalog.py b/services/bench/common/bench_common/model_catalog.py
index 265cc38..c0179eb 100644
--- a/services/bench/common/bench_common/model_catalog.py
+++ b/services/bench/common/bench_common/model_catalog.py
@@ -22,6 +22,10 @@
     "gemini/gemini-3.5-flash",
     "gemini/gemini-flash-latest",
     "gemini/gemini-flash-lite-latest",
+    # Frontier models benched ad hoc via scripts/run_benchmark_matrix.py. Kept
+    # out of FULL_BENCH_MODELS so the canonical full-bench set stays at five.
+    "anthropic/claude-opus-4-8",
+    "openai/gpt-5.5",
 )
 
 ALLOWED_MODELS: tuple[str, ...] = FULL_BENCH_MODELS + EXTRA_MODEL_ALIASES
@@ -37,4 +41,6 @@
     "gemini/gemini-3.5-flash": "Gemini 3.5 Flash",
     "gemini/gemini-flash-latest": "Gemini Flash (latest)",
     "gemini/gemini-flash-lite-latest": "Gemini Flash Lite (latest)",
+    "anthropic/claude-opus-4-8": "Claude Opus 4.8",
+    "openai/gpt-5.5": "GPT-5.5",
 }
diff --git a/services/bench/common/pyproject.toml b/services/bench/common/pyproject.toml
index df48241..2a8dd07 100644
--- a/services/bench/common/pyproject.toml
+++ b/services/bench/common/pyproject.toml
@@ -25,9 +25,11 @@ build-backend = "hatchling.build"
 
 [tool.hatch.build.targets.wheel]
 packages = ["bench_common"]
-
-[tool.hatch.build.targets.wheel.force-include]
-"bench_common/cli/templates" = "bench_common/cli/templates"
+# NOTE: do not add a force-include for bench_common/cli/templates — `packages`
+# already ships every file under bench_common/ (including non-.py template
+# files), so force-including the same paths makes hatchling add each file
+# twice and fail the build ("A second file is being added ... at the same
+# path"). See git history for the duplicate-inclusion regression.
 
 [tool.ruff]
 target-version = "py311"