diff --git a/README.md b/README.md
index bfaeae7d6e..0d21cd8b85 100644
--- a/README.md
+++ b/README.md
@@ -302,7 +302,7 @@ For detailed documentation, see the [TileIR Backend Guide](docs/tileir_backend.m
 
 ## CuTe Backend (CUDA 13+)
 
-Helion has an experimental [CUTLASS CuTe DSL](https://github.com/NVIDIA/cutlass) backend that lowers kernels through CuTe instead of Triton.
+Helion has a beta [CUTLASS CuTe DSL](https://github.com/NVIDIA/cutlass) backend that lowers kernels through CuTe instead of Triton.
 
 **Requirements:** PyTorch must be built against **CUDA 13 or later** (`torch.version.cuda >= "13"`). The backend will refuse to launch on older CUDA builds.
 
diff --git a/docs/installation.md b/docs/installation.md
index 864a0fabcc..2264107be2 100644
--- a/docs/installation.md
+++ b/docs/installation.md
@@ -177,7 +177,7 @@ make html
 
 ### CuTe Backend (CUDA 13+)
 
-To use the experimental [CUTLASS CuTe DSL](https://github.com/NVIDIA/cutlass)
+To use the beta [CUTLASS CuTe DSL](https://github.com/NVIDIA/cutlass)
 backend (`HELION_BACKEND=cute`), PyTorch must be built against **CUDA 13 or
 later** (`torch.version.cuda >= "13"`). The backend refuses to launch on older
 CUDA builds.
diff --git a/helion/_compiler/backend.py b/helion/_compiler/backend.py
index 00b93b0e11..ddbb14c8f1 100644
--- a/helion/_compiler/backend.py
+++ b/helion/_compiler/backend.py
@@ -3,6 +3,7 @@
 import abc
 import collections
 import dataclasses
+import enum
 import functools
 import logging
 import math
@@ -179,6 +180,14 @@ def _validate_subscript_indices(index: list[object]) -> int:
     return narrowed
 
 
+class BackendState(enum.Enum):
+    """Maturity of a codegen backend."""
+
+    EXPERIMENTAL = "experimental"  # Warns on use
+    BETA = "beta"  # No warning, but no backwards-compat guarantee
+    STABLE = "stable"
+
+
 class Backend(abc.ABC):
     """Abstract base class for Helion code generation backends.
 
@@ -195,9 +204,9 @@ def name(self) -> str:
         ...
 
     @property
-    def experimental(self) -> bool:
-        """Whether this backend is experimental and should emit a warning."""
-        return True
+    def state(self) -> BackendState:
+        """Maturity of this backend; experimental backends emit a warning."""
+        return BackendState.EXPERIMENTAL
 
     @property
     def supports_debug_dtype_asserts(self) -> bool:
diff --git a/helion/_compiler/compile_environment.py b/helion/_compiler/compile_environment.py
index 462b83ec34..3c95edeadb 100644
--- a/helion/_compiler/compile_environment.py
+++ b/helion/_compiler/compile_environment.py
@@ -40,6 +40,7 @@
 from .._compat import target_device_capability
 from .._utils import triton_is_available
 from ..language.constexpr import ConstExpr
+from .backend import BackendState
 from .backend_registry import get_backend_class
 from .source_location import SourceLocation
 from .source_location import current_location
@@ -450,7 +451,7 @@ def __init__(
                 "debug_dtype_asserts=True (it emits Triton tl.cast / "
                 "tl.static_assert checks)",
             )
-        if self._backend.experimental:
+        if self._backend.state is BackendState.EXPERIMENTAL:
             from torch._dynamo.utils import warn_once
 
             warn_once(
diff --git a/helion/_compiler/cute/backend.py b/helion/_compiler/cute/backend.py
index b4b6f1b231..b095e93034 100644
--- a/helion/_compiler/cute/backend.py
+++ b/helion/_compiler/cute/backend.py
@@ -23,6 +23,7 @@
 
 from ... import exc
 from ..backend import Backend
+from ..backend import BackendState
 from ..backend import _active_loop_block_ids
 from ..backend import _attention_flash_gate_enabled
 from ..backend import _attention_flash_supported
@@ -1409,6 +1410,10 @@ def codegen_config(self, config: Config) -> Config:
     def name(self) -> str:
         return "cute"
 
+    @property
+    def state(self) -> BackendState:
+        return BackendState.BETA
+
     @property
     def supports_eager_prepared_call(self) -> bool:
         return True
diff --git a/helion/_compiler/flydsl/backend.py b/helion/_compiler/flydsl/backend.py
index 55a80ed65a..e320caa40d 100644
--- a/helion/_compiler/flydsl/backend.py
+++ b/helion/_compiler/flydsl/backend.py
@@ -14,6 +14,7 @@
 from ... import exc
 from ..ast_extension import expr_from_string
 from ..backend import Backend
+from ..backend import BackendState
 
 if TYPE_CHECKING:
     import ast
@@ -120,8 +121,8 @@ def name(self) -> str:
         return "flydsl"
 
     @property
-    def experimental(self) -> bool:
-        return True
+    def state(self) -> BackendState:
+        return BackendState.EXPERIMENTAL
 
     def validate_environment(self) -> None:
         try:
diff --git a/helion/_compiler/metal/backend.py b/helion/_compiler/metal/backend.py
index e3c4478619..f79031abab 100644
--- a/helion/_compiler/metal/backend.py
+++ b/helion/_compiler/metal/backend.py
@@ -12,6 +12,7 @@
 
 from ... import exc
 from ..backend import Backend
+from ..backend import BackendState
 from ..backend import _largest_divisor_at_most
 from ..cute.backend import CuteBackend
 from ._constants import MAX_THREAD_AXES
@@ -78,6 +79,10 @@ def _get_dtype_to_metal() -> dict[torch.dtype, str]:
     def name(self) -> str:
         return "metal"
 
+    @property
+    def state(self) -> BackendState:
+        return BackendState.EXPERIMENTAL
+
     def dtype_str(self, dtype: torch.dtype) -> str:
         dtype_map = self._get_dtype_to_metal()
         if dtype not in dtype_map:
diff --git a/helion/_compiler/pallas/backend.py b/helion/_compiler/pallas/backend.py
index 1cfbb32a87..24e32ed8b6 100644
--- a/helion/_compiler/pallas/backend.py
+++ b/helion/_compiler/pallas/backend.py
@@ -20,6 +20,7 @@
 from ..ast_extension import expr_from_string
 from ..backend import AutotuneGridPolicy
 from ..backend import Backend
+from ..backend import BackendState
 from ..backend import LauncherInfo
 from ..backend import _loop_contains_matmul
 from ..backend import _validate_subscript_indices
@@ -274,6 +275,10 @@ class PallasBackend(Backend):
     def name(self) -> str:
         return "pallas"
 
+    @property
+    def state(self) -> BackendState:
+        return BackendState.BETA
+
     @staticmethod
     # Overrides Backend.map_dot_precision.
     def map_dot_precision(precision: DotPrecision) -> str:
diff --git a/helion/_compiler/triton/backend.py b/helion/_compiler/triton/backend.py
index 871b066bd5..c8c5d6c6a0 100644
--- a/helion/_compiler/triton/backend.py
+++ b/helion/_compiler/triton/backend.py
@@ -18,6 +18,7 @@
 from ... import _compat
 from ... import exc
 from ..backend import Backend
+from ..backend import BackendState
 from ..backend import LauncherInfo
 from ..backend import log
 
@@ -61,8 +62,8 @@ def name(self) -> str:
         return "triton"
 
     @property
-    def experimental(self) -> bool:
-        return False
+    def state(self) -> BackendState:
+        return BackendState.STABLE
 
     @property
     def supports_debug_dtype_asserts(self) -> bool:
diff --git a/test/test_backend_registry.py b/test/test_backend_registry.py
index 361cbda2aa..366e345f58 100644
--- a/test/test_backend_registry.py
+++ b/test/test_backend_registry.py
@@ -11,6 +11,7 @@
 from helion._compiler import backend_registry
 from helion._compiler.aten_lowering import AtenLowering
 from helion._compiler.backend import Backend
+from helion._compiler.backend import BackendState
 from helion._compiler.backend_registry import _REGISTRY
 from helion._compiler.backend_registry import all_reserved_launch_param_names
 from helion._compiler.backend_registry import get_backend_class
@@ -90,8 +91,8 @@ def library_imports(self) -> dict[str, str]:
         try:
             self.assertIn("_test_custom", list_backends())
             self.assertIs(get_backend_class("_test_custom"), _TestBackend)
-            # custom backends default to experimental=True
-            self.assertTrue(_TestBackend().experimental)
+            # custom backends default to experimental
+            self.assertIs(_TestBackend().state, BackendState.EXPERIMENTAL)
         finally:
             _REGISTRY.pop("_test_custom", None)