fix(anthropic, bedrock): drop temperature/top_p for Claude Opus 4.7

Anthropic's Messages API returns 400
(``temperature is deprecated for this model``) when ``temperature``,
``top_p``, or ``top_k`` is set to any non-default value on Claude
Opus 4.7 — per the Opus 4.7 migration guide. Previously
``AnthropicConfig.get_supported_openai_params`` unconditionally
advertised ``temperature`` and ``top_p`` for every Anthropic model,
so ``litellm.drop_params=True`` was a no-op on 4.7: litellm had no
reason to drop a param it believed was supported, the value was
forwarded to the API, and end users took a hard 400. The same shape
hits ``AmazonConverseConfig`` for Bedrock-routed Opus 4.7.

Fix follows the maintainers' existing capability-flag pattern
(``supports_*`` keys on model_prices_and_context_window.json read via
``_is_explicitly_disabled_factory``) rather than hardcoding model
strings in transformation logic:

* Mark every Anthropic-API-shape Opus 4.7 entry with
  ``supports_temperature: false`` / ``supports_top_p: false`` —
  canonical Anthropic, all four Bedrock Converse regions
  (anthropic/global/us/eu/au), Azure AI, and Vertex AI. Sonnet 4.6,
  Haiku 4.5, Opus 4.6, and 3.x models are unchanged. OpenRouter and
  Perplexity entries are deliberately left alone — they route through
  different configs.

* New ``AnthropicConfig._param_explicitly_unsupported`` helper reads
  ``supports_{param}`` via the shared
  ``_is_explicitly_disabled_factory`` (same chain that drives
  ``_is_reasoning_effort_level_explicitly_disabled`` in
  ``OpenAIGPT5Config``). When the registry lookup misses — e.g. an
  unreleased dated 4.7 snapshot — falls back to the existing
  ``_is_claude_4_7_model`` family check so future 4.7 variants are
  covered before anyone updates the JSON.

* ``AnthropicConfig.get_supported_openai_params`` filters the helper
  across every param (not just temperature) so future deprecations
  only need a JSON entry. ``map_openai_params`` honours the same
  helper inside the ``temperature`` / ``top_p`` branches so direct
  callers don't leak the params either, even without ``drop_params``.

* ``AmazonConverseConfig.get_supported_openai_params`` mirrors the
  filter so Bedrock-routed Opus 4.7 gets the same treatment.

The Databricks workspace endpoint surfaces the same Anthropic
deprecation (see the issue's follow-up comment); leaving that to a
separate PR keeps this change focused on the two configs the bug
report names.

The backup ``litellm/model_prices_and_context_window_backup.json``
absorbs upstream drift from earlier unrelated merges so the
``ci_cd/check_files_match.py`` byte-identity check passes — same
constraint that hit #26246.

Tests:

* New parametrized unit tests assert ``temperature`` / ``top_p`` are
  filtered out of ``AnthropicConfig.get_supported_openai_params`` for
  ``claude-opus-4-7``, the dated variant
  ``claude-opus-4-7-20260416``, and the vendor-prefixed
  ``anthropic/claude-opus-4-7``.
* Family-fallback test covers an unreleased dated snapshot not in the
  registry.
* Regression tests pin Sonnet 4.6 (both dated and alias), Haiku 4.5,
  Sonnet 3.5, Sonnet 3.7, and Opus 4.6 to still advertise both
  sampling params.
* ``map_openai_params`` test confirms explicit ``temperature=0.3,
  top_p=0.9`` does not reach ``optional_params`` for Opus 4.7 but
  does for Opus 4.6.
* End-to-end-shape test reproduces the issue: ``drop_params=True`` +
  ``temperature=0.1`` + ``transform_request`` no longer leaks
  ``temperature`` into the Anthropic request body.
* Bedrock Converse mirror suite covers all four region variants and
  pins Sonnet 4.6 / Haiku 4.5 / 3.5-sonnet on Bedrock as regressions.

Existing Anthropic + Bedrock test suites stay green
(``pytest tests/test_litellm/llms/anthropic
tests/test_litellm/llms/bedrock/chat`` — 976 passed locally).

Fixes #26444

litellm/llms/anthropic/chat/transformation.py

@@ -80,6 +80,7 @@ from litellm.types.utils import (
 from litellm.utils import (
     ModelResponse,
     Usage,
+    _is_explicitly_disabled_factory,
     _supports_factory,
     add_dummy_tool,
     any_assistant_message_has_thinking_blocks,
@@ -431,6 +432,34 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig):
             llm_provider=llm_provider,
         )
 
+    @staticmethod
+    def _param_explicitly_unsupported(model: str, param: str) -> bool:
+        """Whether the upstream API rejects ``param`` for ``model``.
+
+        Primary signal is the model map: ``supports_{param}: false`` flips
+        this to True via the same ``_is_explicitly_disabled_factory``
+        chain other capability gates use, so a missing flag is treated
+        as "supported".
+
+        Falls back to a model-family check for the Opus 4.7 sampling
+        deprecation (``temperature`` / ``top_p`` / ``top_k`` return 400
+        on the Anthropic Messages API per the Opus 4.7 migration guide).
+        This keeps unreleased dated 4.7 snapshots and any
+        provider-prefixed alias that isn't yet in the JSON covered.
+        """
+        try:
+            if _is_explicitly_disabled_factory(
+                model=model,
+                custom_llm_provider=None,
+                key=f"supports_{param}",
+            ):
+                return True
+        except Exception:
+            pass
+        if param in ("temperature", "top_p", "top_k"):
+            return AnthropicConfig._is_claude_4_7_model(model)
+        return False
+
     def get_supported_openai_params(self, model: str):
         params = [
             "stream",
@@ -463,6 +492,16 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig):
             params.append("thinking")
             params.append("reasoning_effort")
 
+        # Strip sampling params the upstream API rejects for this model.
+        # Anthropic returns a 400 for ``temperature`` / ``top_p`` / ``top_k``
+        # on Claude Opus 4.7; the model map encodes this via
+        # ``supports_temperature: false`` / ``supports_top_p: false``.
+        params = [
+            p
+            for p in params
+            if not AnthropicConfig._param_explicitly_unsupported(model, p)
+        ]
+
         return params
 
     @staticmethod
@@ -1452,8 +1491,15 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig):
                 if _value is not None:
                     optional_params["stop_sequences"] = _value
             elif param == "temperature":
+                if AnthropicConfig._param_explicitly_unsupported(model, "temperature"):
+                    # Anthropic Messages API rejects ``temperature`` for
+                    # Opus 4.7; silently drop rather than forwarding the
+                    # value to a guaranteed 400.
+                    continue
                 optional_params["temperature"] = value
             elif param == "top_p":
+                if AnthropicConfig._param_explicitly_unsupported(model, "top_p"):
+                    continue
                 optional_params["top_p"] = value
             elif param == "response_format" and isinstance(value, dict):
                 if any(

litellm/llms/bedrock/chat/converse_transformation.py

@@ -573,6 +573,18 @@ class AmazonConverseConfig(BaseConfig):
         ):
             supported_params.append("thinking")
             supported_params.append("reasoning_effort")
+
+        # Strip sampling params the upstream API rejects for this model.
+        # Anthropic's Messages API (surfaced via Bedrock Converse for
+        # claude-opus-4-7) returns 400 for ``temperature`` / ``top_p``;
+        # the model map encodes this via ``supports_temperature: false``
+        # / ``supports_top_p: false`` so ``drop_params=True`` strips
+        # them before the request leaves litellm.
+        supported_params = [
+            p
+            for p in supported_params
+            if not AnthropicConfig._param_explicitly_unsupported(model, p)
+        ]
         return supported_params
 
     def map_tool_choice_values(

litellm/model_prices_and_context_window_backup.json

@@ -1155,6 +1155,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -1201,6 +1203,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -1232,6 +1236,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -1262,6 +1268,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -1292,6 +1300,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -1448,6 +1458,35 @@
         "supports_native_structured_output": true,
         "supports_minimal_reasoning_effort": true
     },
+    "jp.anthropic.claude-sonnet-4-6": {
+        "cache_creation_input_token_cost": 4.125e-06,
+        "cache_read_input_token_cost": 3.3e-07,
+        "input_cost_per_token": 3.3e-06,
+        "litellm_provider": "bedrock_converse",
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 64000,
+        "max_tokens": 64000,
+        "mode": "chat",
+        "output_cost_per_token": 1.65e-05,
+        "search_context_cost_per_query": {
+            "search_context_size_high": 0.01,
+            "search_context_size_low": 0.01,
+            "search_context_size_medium": 0.01
+        },
+        "supports_assistant_prefill": true,
+        "supports_computer_use": true,
+        "supports_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_max_reasoning_effort": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "tool_use_system_prompt_tokens": 346,
+        "supports_native_structured_output": true,
+        "supports_minimal_reasoning_effort": true
+    },
     "anthropic.claude-sonnet-4-20250514-v1:0": {
         "cache_creation_input_token_cost": 3.75e-06,
         "cache_read_input_token_cost": 3e-07,
@@ -1993,6 +2032,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -9602,6 +9643,7 @@
             "search_context_size_low": 0.01,
             "search_context_size_medium": 0.01
         },
+        "supports_adaptive_thinking": true,
         "supports_assistant_prefill": true,
         "supports_computer_use": true,
         "supports_function_calling": true,
@@ -9795,6 +9837,7 @@
             "search_context_size_low": 0.01,
             "search_context_size_medium": 0.01
         },
+        "supports_adaptive_thinking": true,
         "supports_assistant_prefill": false,
         "supports_computer_use": true,
         "supports_function_calling": true,
@@ -9828,6 +9871,7 @@
             "search_context_size_low": 0.01,
             "search_context_size_medium": 0.01
         },
+        "supports_adaptive_thinking": true,
         "supports_assistant_prefill": false,
         "supports_computer_use": true,
         "supports_function_calling": true,
@@ -9861,6 +9905,7 @@
             "search_context_size_low": 0.01,
             "search_context_size_medium": 0.01
         },
+        "supports_adaptive_thinking": true,
         "supports_assistant_prefill": false,
         "supports_computer_use": true,
         "supports_function_calling": true,
@@ -9868,6 +9913,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -9895,6 +9942,7 @@
             "search_context_size_low": 0.01,
             "search_context_size_medium": 0.01
         },
+        "supports_adaptive_thinking": true,
         "supports_assistant_prefill": false,
         "supports_computer_use": true,
         "supports_function_calling": true,
@@ -9902,6 +9950,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -32957,6 +33007,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -32986,6 +33038,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,

model_prices_and_context_window.json

@@ -1155,6 +1155,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -1201,6 +1203,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -1232,6 +1236,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -1262,6 +1268,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -1292,6 +1300,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -2022,6 +2032,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -9901,6 +9913,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -9936,6 +9950,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -32991,6 +33007,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
@@ -33020,6 +33038,8 @@
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
+        "supports_temperature": false,
+        "supports_top_p": false,
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,

tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_transformation.py

@@ -2350,6 +2350,154 @@ def test_get_supported_params_includes_reasoning_for_sonnet_4_6_dotted_alias():
     assert "reasoning_effort" in params
 
 
+# --- Opus 4.7 ``temperature`` / ``top_p`` deprecation (issue #26444) ---
+#
+# Anthropic's Messages API returns 400 (``temperature is deprecated for this
+# model``) when ``temperature`` or ``top_p`` is sent to Claude Opus 4.7 with a
+# non-default value. The model registry encodes this as
+# ``supports_temperature: false`` / ``supports_top_p: false`` on every Opus 4.7
+# entry (Anthropic, Bedrock Converse, Vertex, Azure AI), and
+# ``AnthropicConfig`` reads those flags via ``_is_explicitly_disabled_factory``
+# to filter the supported-params list and drop the values inside
+# ``map_openai_params``. A static fallback on ``_is_claude_4_7_model`` covers
+# unreleased dated snapshots that aren't in the registry yet.
+
+
+@pytest.mark.parametrize(
+    "model",
+    [
+        "claude-opus-4-7",
+        "claude-opus-4-7-20260416",
+        "anthropic/claude-opus-4-7",
+    ],
+)
+def test_opus_4_7_drops_temperature_and_top_p_from_supported_params(
+    monkeypatch, model
+):
+    """Opus 4.7 must not advertise ``temperature`` / ``top_p`` so ``drop_params=True`` strips them."""
+    monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
+    # Force a clean reload so the local JSON is honoured for the assertion.
+    import importlib
+
+    import litellm as _litellm
+
+    importlib.reload(_litellm)
+
+    config = AnthropicConfig()
+    params = config.get_supported_openai_params(model=model)
+
+    assert "temperature" not in params, (
+        f"temperature should be filtered from Opus 4.7 supported params; got {params!r}"
+    )
+    assert "top_p" not in params, (
+        f"top_p should be filtered from Opus 4.7 supported params; got {params!r}"
+    )
+
+
+def test_opus_4_7_unknown_dated_variant_falls_back_to_family_check(monkeypatch):
+    """Dated Opus 4.7 snapshots not yet in the model registry are still covered
+    by the ``_is_claude_4_7_model`` family fallback."""
+    monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
+    config = AnthropicConfig()
+
+    # Plausible future snapshot id not present in model_prices_and_context_window.json.
+    params = config.get_supported_openai_params(model="claude-opus-4-7-20991231")
+
+    assert "temperature" not in params
+    assert "top_p" not in params
+
+
+@pytest.mark.parametrize(
+    "model",
+    [
+        "claude-sonnet-4-6",
+        "claude-sonnet-4-6-20260219",
+        "claude-haiku-4-5",
+        "claude-3-5-sonnet-20241022",
+        "claude-3-7-sonnet-20250219",
+        "claude-opus-4-6",
+    ],
+)
+def test_non_opus_4_7_models_still_support_temperature_and_top_p(
+    monkeypatch, model
+):
+    """Regression guard: only Opus 4.7 deprecated temperature; older models
+    (including same-generation 4.6 and reasoning-family 3.7-sonnet) must keep
+    advertising ``temperature`` and ``top_p`` so existing call sites keep working."""
+    monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
+    config = AnthropicConfig()
+    params = config.get_supported_openai_params(model=model)
+
+    assert "temperature" in params, (
+        f"temperature should remain supported for {model}; got {params!r}"
+    )
+    assert "top_p" in params, (
+        f"top_p should remain supported for {model}; got {params!r}"
+    )
+
+
+def test_opus_4_7_map_openai_params_drops_temperature_and_top_p(monkeypatch):
+    """``map_openai_params`` must not leak ``temperature`` / ``top_p`` into the
+    Anthropic request body for Opus 4.7, even when callers pass them
+    explicitly without ``drop_params=True`` (defense in depth)."""
+    monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
+    config = AnthropicConfig()
+
+    optional_params = config.map_openai_params(
+        non_default_params={"temperature": 0.3, "top_p": 0.9},
+        optional_params={},
+        model="claude-opus-4-7",
+        drop_params=False,
+    )
+
+    assert "temperature" not in optional_params
+    assert "top_p" not in optional_params
+
+
+def test_opus_4_6_map_openai_params_preserves_temperature_and_top_p(monkeypatch):
+    """Regression guard on the mapping site: Opus 4.6 still receives both
+    sampling params verbatim."""
+    monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
+    config = AnthropicConfig()
+
+    optional_params = config.map_openai_params(
+        non_default_params={"temperature": 0.3, "top_p": 0.9},
+        optional_params={},
+        model="claude-opus-4-6",
+        drop_params=False,
+    )
+
+    assert optional_params["temperature"] == 0.3
+    assert optional_params["top_p"] == 0.9
+
+
+def test_opus_4_7_drop_params_true_strips_temperature_before_transform(monkeypatch):
+    """End-to-end-shape check that mirrors issue #26444's reproduction: a
+    ``drop_params=True`` call with ``temperature`` set must produce a
+    transformed Anthropic request body that does not contain ``temperature``."""
+    monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
+    config = AnthropicConfig()
+    messages = [{"role": "user", "content": "hi"}]
+
+    optional_params = config.map_openai_params(
+        non_default_params={"temperature": 0.1},
+        optional_params={},
+        model="claude-opus-4-7",
+        drop_params=True,
+    )
+    body = config.transform_request(
+        model="claude-opus-4-7",
+        messages=messages,
+        optional_params=optional_params,
+        litellm_params={},
+        headers={},
+    )
+
+    assert "temperature" not in body, (
+        f"temperature leaked into the Anthropic request body: {body!r}"
+    )
+
+
 def test_sonnet_4_6_reasoning_effort_to_transform_request_payload():
     """
     Sonnet 4.6 should convert reasoning_effort to adaptive thinking in final request payload.

tests/test_litellm/llms/bedrock/chat/test_converse_transformation.py

@@ -4578,3 +4578,69 @@ def test_transform_response_does_not_leak_body_on_parse_failure():
     msg = str(exc_info.value)
     assert "secret content" not in msg
     assert "Error converting to valid response block" in msg
+
+
+# --- Opus 4.7 ``temperature`` / ``top_p`` deprecation (issue #26444) ---
+#
+# Bedrock Converse surfaces the Anthropic Messages API for ``anthropic.claude-*``
+# models, so the same 400 (``temperature is deprecated for this model``) hits
+# Opus 4.7 on Bedrock. ``AmazonConverseConfig.get_supported_openai_params``
+# reads the ``supports_temperature`` / ``supports_top_p`` flags off the model
+# registry via ``AnthropicConfig._param_explicitly_unsupported`` so
+# ``drop_params=True`` strips them before the call leaves litellm.
+
+
+@pytest.mark.parametrize(
+    "model",
+    [
+        "bedrock/converse/anthropic.claude-opus-4-7",
+        "bedrock/converse/us.anthropic.claude-opus-4-7",
+        "bedrock/converse/eu.anthropic.claude-opus-4-7",
+        "bedrock/converse/global.anthropic.claude-opus-4-7",
+    ],
+)
+def test_bedrock_converse_opus_4_7_drops_temperature_and_top_p(monkeypatch, model):
+    """Bedrock-routed Opus 4.7 must not advertise ``temperature`` / ``top_p``."""
+    monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
+    import importlib
+
+    import litellm as _litellm
+
+    importlib.reload(_litellm)
+
+    config = AmazonConverseConfig()
+    params = config.get_supported_openai_params(model=model)
+
+    assert "temperature" not in params, (
+        f"temperature should be filtered for {model}; got {params!r}"
+    )
+    assert "top_p" not in params, (
+        f"top_p should be filtered for {model}; got {params!r}"
+    )
+
+
+@pytest.mark.parametrize(
+    "model",
+    [
+        "bedrock/converse/anthropic.claude-sonnet-4-6",
+        "bedrock/converse/us.anthropic.claude-opus-4-6",
+        "bedrock/converse/anthropic.claude-haiku-4-5-20251001-v1:0",
+        "bedrock/converse/anthropic.claude-3-5-sonnet-20241022-v2:0",
+    ],
+)
+def test_bedrock_converse_non_opus_4_7_models_keep_temperature_and_top_p(
+    monkeypatch, model
+):
+    """Regression guard: same-generation Sonnet 4.6 / Haiku 4.5 / older 3.5
+    models on Bedrock Converse must continue to advertise ``temperature``
+    and ``top_p``."""
+    monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
+    config = AmazonConverseConfig()
+    params = config.get_supported_openai_params(model=model)
+
+    assert "temperature" in params, (
+        f"temperature should remain supported for {model}; got {params!r}"
+    )
+    assert "top_p" in params, (
+        f"top_p should remain supported for {model}; got {params!r}"
+    )