fix(bedrock): gate in-place system role messages on model support for Claude Invoke

litellm/litellm_core_utils/fallback_generalizations.py

@@ -8,9 +8,12 @@
 routing, and the remaining fields (``mode``, ``supports_*``, context window,
 pricing, ...) drive ``get_model_info`` / ``supports_*``.
 
-Precedence: rules are evaluated in file order and the first match wins. They are
-consulted only after exact and case-insensitive lookups miss, so an exact entry
-always takes precedence over a rule.
+Precedence: rules are evaluated in file order and the first match wins. Callers
+with extra constraints (model-info resolution checks the provider) use
+``match_all_fallback_generalizations`` to skip inapplicable earlier rules instead
+of discarding the model name. Rules are consulted only after exact and
+case-insensitive lookups miss, so an exact entry always takes precedence over a
+rule.
 
 Patterns are matched case-insensitively with ``re.search`` and are not implicitly
 anchored: a rule must include ``^`` and ``$`` (as the shipped rules do) to bind to
@@ -105,15 +108,15 @@ def _compile(self) -> list[tuple[re.Pattern, dict]]:
                 )
         return compiled
 
-    def match(self, model: str) -> Optional[dict]:
+    def matches(self, model: str) -> list[dict]:
         if not model:
-            return None
+            return []
         if self._compiled is None:
             self._compiled = self._compile()
-        for pattern, model_info in self._compiled:
-            if pattern.search(model) is not None:
-                return dict(model_info)
-        return None
+        return [dict(model_info) for pattern, model_info in self._compiled if pattern.search(model) is not None]
+
+    def match(self, model: str) -> Optional[dict]:
+        return next(iter(self.matches(model)), None)
 
 
 _registry = _FallbackGeneralizations()
@@ -139,3 +142,12 @@ def match_fallback_generalization(model: str) -> Optional[dict]:
     O(number of rules). Only call this once exact lookups have missed.
     """
     return _registry.match(model)
+
+
+def match_all_fallback_generalizations(model: str) -> list[dict]:
+    """Return the ``model_info`` of every rule whose regex matches ``model``, in rule order.
+
+    Lets a caller with extra constraints (e.g. a provider match) skip an
+    inapplicable earlier rule instead of discarding the whole candidate.
+    """
+    return _registry.matches(model)

litellm/llms/bedrock/messages/invoke_transformations/anthropic_claude3_transformation.py

@@ -93,31 +93,48 @@ def _as_system_content_blocks(value: Any) -> list[Any]:
             return [{"type": "text", "text": value}]
         return [value]
 
-    def _normalize_system_role_messages_for_bedrock(self, anthropic_messages_request: dict) -> None:
-        """Bedrock Invoke rejects a conversation that opens with ``role: "system"``
-        entries inside ``messages`` ("messages.0: use the top-level 'system'
-        parameter for the initial system prompt"); Anthropic Messages carries that
-        content in the top-level ``system`` field, so hoist the leading run of
-        system entries there. Mid-conversation system entries (e.g. Claude Code's
-        ``mid-conversation-system-2026-04-07`` reminders) are accepted by Invoke in
-        place and MUST stay in place: hoisting one mutates the ``system`` prefix
-        and invalidates the prompt cache for the entire message history.
+    @staticmethod
+    def _is_system_role_message(message: Any) -> bool:
+        return isinstance(message, dict) and message.get("role") == "system"
+
+    def _normalize_system_role_messages_for_bedrock(self, anthropic_messages_request: dict, model: str) -> None:
+        """Bedrock Invoke validates ``role: "system"`` entries inside ``messages``
+        per model. Models carrying ``supports_mid_conversation_system`` in the
+        cost map (the Opus 4.8 family) only reject a leading run ("messages.0:
+        use the top-level 'system' parameter for the initial system prompt") and
+        accept mid-conversation entries (e.g. Claude Code's
+        ``mid-conversation-system-2026-04-07`` reminders) in place, where they
+        MUST stay: hoisting one mutates the ``system`` prefix and invalidates the
+        prompt cache for the entire message history. Older Claude models (Opus
+        4.7, Sonnet 4.6, Haiku 4.5, ...) reject the role in every position
+        ("role 'system' is not supported on this model"), so without the flag
+        every system entry is hoisted into the top-level ``system`` field.
         Billing-header system blocks are stripped from the top-level ``system``
         field regardless of whether anything was hoisted."""
         messages = anthropic_messages_request.get("messages")
         if not isinstance(messages, list):
             return
-        leading_count = next(
-            (i for i, m in enumerate(messages) if not (isinstance(m, dict) and m.get("role") == "system")),
-            len(messages),
-        )
-        if leading_count:
-            anthropic_messages_request["messages"] = messages[leading_count:]
+        if _supports_factory(
+            model=model,
+            custom_llm_provider="bedrock",
+            key="supports_mid_conversation_system",
+        ):
+            leading_count = next(
+                (i for i, m in enumerate(messages) if not self._is_system_role_message(m)),
+                len(messages),
+            )
+            hoisted = messages[:leading_count]
+            remaining = messages[leading_count:]
+        else:
+            hoisted = [m for m in messages if self._is_system_role_message(m)]
+            remaining = [m for m in messages if not self._is_system_role_message(m)]
+        if hoisted:
+            anthropic_messages_request["messages"] = remaining
         system_content = [
             block
             for source in (
                 anthropic_messages_request.get("system"),
-                *(m.get("content") for m in messages[:leading_count]),
+                *(m.get("content") for m in hoisted),
             )
             for block in self._as_system_content_blocks(source)
         ]
@@ -674,7 +691,7 @@ def transform_anthropic_messages_request(
             litellm_params=litellm_params,
             headers=headers,
         )
-        self._normalize_system_role_messages_for_bedrock(anthropic_messages_request)
+        self._normalize_system_role_messages_for_bedrock(anthropic_messages_request, model=model)
         #########################################################
         ############## BEDROCK Invoke SPECIFIC TRANSFORMATION ###
         #########################################################

litellm/model_prices_and_context_window_backup.json

@@ -1481,6 +1481,7 @@
     "anthropic.claude-opus-4-8": {
         "bedrock_converse_supports_strict_tools": false,
         "supports_adaptive_thinking": true,
+        "supports_mid_conversation_system": true,
         "cache_creation_input_token_cost": 6.25e-06,
         "cache_creation_input_token_cost_above_1hr": 1e-05,
         "cache_read_input_token_cost": 5e-07,
@@ -1516,6 +1517,7 @@
     "global.anthropic.claude-opus-4-8": {
         "bedrock_converse_supports_strict_tools": false,
         "supports_adaptive_thinking": true,
+        "supports_mid_conversation_system": true,
         "cache_creation_input_token_cost": 6.25e-06,
         "cache_creation_input_token_cost_above_1hr": 1e-05,
         "cache_read_input_token_cost": 5e-07,
@@ -1551,6 +1553,7 @@
     "us.anthropic.claude-opus-4-8": {
         "bedrock_converse_supports_strict_tools": false,
         "supports_adaptive_thinking": true,
+        "supports_mid_conversation_system": true,
         "cache_creation_input_token_cost": 6.875e-06,
         "cache_creation_input_token_cost_above_1hr": 1.1e-05,
         "cache_read_input_token_cost": 5.5e-07,
@@ -1586,6 +1589,7 @@
     "eu.anthropic.claude-opus-4-8": {
         "bedrock_converse_supports_strict_tools": false,
         "supports_adaptive_thinking": true,
+        "supports_mid_conversation_system": true,
         "cache_creation_input_token_cost": 6.875e-06,
         "cache_creation_input_token_cost_above_1hr": 1.1e-05,
         "cache_read_input_token_cost": 5.5e-07,
@@ -1621,6 +1625,7 @@
     "au.anthropic.claude-opus-4-8": {
         "bedrock_converse_supports_strict_tools": false,
         "supports_adaptive_thinking": true,
+        "supports_mid_conversation_system": true,
         "cache_creation_input_token_cost": 6.875e-06,
         "cache_creation_input_token_cost_above_1hr": 1.1e-05,
         "cache_read_input_token_cost": 5.5e-07,
@@ -44998,6 +45003,17 @@
     },
     "fallback_generalizations": {
         "rules": [
+            {
+                "name": "bedrock-anthropic-claude-mid-conversation-system",
+                "pattern": "anthropic\\.claude-[a-z]+-(?:4[-._](?:[89]|[1-9]\\d)(?!\\d)|(?:[5-9]|[1-9]\\d)(?!\\d)(?:[-._]\\d{1,2}(?!\\d))?)",
+                "description": "Bedrock Invoke ids for Claude 4.8 or higher: anthropic.claude-<family> with minor 4.8 through 4.99, any 5.x or later major-minor, or a bare 5+ major, which also admits new families such as fable. These models accept mid-conversation role system messages in place (verified live on Opus 4.8, Sonnet 5 and Fable 5), so unmapped future Bedrock Claudes keep the cache-preserving in-place handling instead of the hoist-all default. Listed first so bare-id provider inference, which takes the first pattern hit, resolves these Bedrock ids to bedrock; model-info resolution skips provider-mismatched rules either way.",
+                "extends": "anthropic-claude",
+                "model_info": {
+                    "litellm_provider": "bedrock",
+                    "supports_adaptive_thinking": true,
+                    "supports_mid_conversation_system": true
+                }
+            },
             {
                 "name": "anthropic-claude-adaptive-thinking",
                 "pattern": "(?:opus|sonnet|haiku)[-._](?:4[-._](?:[6-9]|[1-9]\\d)(?!\\d)|(?:[5-9]|[1-9]\\d{1,})[-._]\\d{1,2}(?!\\d))",

litellm/types/utils.py

@@ -143,6 +143,7 @@ class ProviderSpecificModelInfo(TypedDict, total=False):
     supports_web_search: Optional[bool]
     supports_reasoning: Optional[bool]
     supports_adaptive_thinking: Optional[bool]
+    supports_mid_conversation_system: Optional[bool]
     supports_url_context: Optional[bool]
     supports_none_reasoning_effort: Optional[bool]
     supports_minimal_reasoning_effort: Optional[bool]

litellm/utils.py

@@ -61,7 +61,7 @@
 )
 from litellm._uuid import uuid
 from litellm.litellm_core_utils.fallback_generalizations import (
-    match_fallback_generalization,
+    match_all_fallback_generalizations,
 )
 from litellm.constants import (
     DEFAULT_CHAT_COMPLETION_PARAM_VALUES,
@@ -5046,9 +5046,10 @@ def _get_model_info_from_generalization(
     """Resolve an unmapped model via a declarative fallback-generalization rule.
 
     Tries the same name candidates as the exact lookups, in the same order, and
-    returns ``(matched_name, model_info)`` for the first candidate whose rule also
-    satisfies the provider constraint. O(number of rules); only call after the
-    exact lookups have missed.
+    returns ``(matched_name, model_info)`` for the first matching rule that also
+    satisfies the provider constraint; a rule scoped to another provider is
+    skipped in favor of later rules rather than discarding the candidate.
+    O(number of rules); only call after the exact lookups have missed.
     """
     candidates = [
         potential_model_names["combined_model_name"],
@@ -5058,11 +5059,9 @@ def _get_model_info_from_generalization(
         potential_model_names["stripped_model_name"],
     ]
     for candidate in candidates:
-        generalized_info = match_fallback_generalization(candidate)
-        if generalized_info is not None and _check_provider_match(
-            model_info=generalized_info, custom_llm_provider=custom_llm_provider
-        ):
-            return candidate, generalized_info
+        for generalized_info in match_all_fallback_generalizations(candidate):
+            if _check_provider_match(model_info=generalized_info, custom_llm_provider=custom_llm_provider):
+                return candidate, generalized_info
     return None
 
 
@@ -5472,6 +5471,7 @@ def _get_model_info_helper(
                 supports_url_context=_model_info.get("supports_url_context", None),
                 supports_reasoning=_model_info.get("supports_reasoning", None),
                 supports_adaptive_thinking=_model_info.get("supports_adaptive_thinking", None),
+                supports_mid_conversation_system=_model_info.get("supports_mid_conversation_system", None),
                 supports_none_reasoning_effort=_model_info.get("supports_none_reasoning_effort", None),
                 supports_minimal_reasoning_effort=_model_info.get("supports_minimal_reasoning_effort", None),
                 supports_low_reasoning_effort=_model_info.get("supports_low_reasoning_effort", None),

model_prices_and_context_window.json

@@ -1481,6 +1481,7 @@
     "anthropic.claude-opus-4-8": {
         "bedrock_converse_supports_strict_tools": false,
         "supports_adaptive_thinking": true,
+        "supports_mid_conversation_system": true,
         "cache_creation_input_token_cost": 6.25e-06,
         "cache_creation_input_token_cost_above_1hr": 1e-05,
         "cache_read_input_token_cost": 5e-07,
@@ -1516,6 +1517,7 @@
     "global.anthropic.claude-opus-4-8": {
         "bedrock_converse_supports_strict_tools": false,
         "supports_adaptive_thinking": true,
+        "supports_mid_conversation_system": true,
         "cache_creation_input_token_cost": 6.25e-06,
         "cache_creation_input_token_cost_above_1hr": 1e-05,
         "cache_read_input_token_cost": 5e-07,
@@ -1551,6 +1553,7 @@
     "us.anthropic.claude-opus-4-8": {
         "bedrock_converse_supports_strict_tools": false,
         "supports_adaptive_thinking": true,
+        "supports_mid_conversation_system": true,
         "cache_creation_input_token_cost": 6.875e-06,
         "cache_creation_input_token_cost_above_1hr": 1.1e-05,
         "cache_read_input_token_cost": 5.5e-07,
@@ -1586,6 +1589,7 @@
     "eu.anthropic.claude-opus-4-8": {
         "bedrock_converse_supports_strict_tools": false,
         "supports_adaptive_thinking": true,
+        "supports_mid_conversation_system": true,
         "cache_creation_input_token_cost": 6.875e-06,
         "cache_creation_input_token_cost_above_1hr": 1.1e-05,
         "cache_read_input_token_cost": 5.5e-07,
@@ -1621,6 +1625,7 @@
     "au.anthropic.claude-opus-4-8": {
         "bedrock_converse_supports_strict_tools": false,
         "supports_adaptive_thinking": true,
+        "supports_mid_conversation_system": true,
         "cache_creation_input_token_cost": 6.875e-06,
         "cache_creation_input_token_cost_above_1hr": 1.1e-05,
         "cache_read_input_token_cost": 5.5e-07,
@@ -45231,6 +45236,17 @@
     },
     "fallback_generalizations": {
         "rules": [
+            {
+                "name": "bedrock-anthropic-claude-mid-conversation-system",
+                "pattern": "anthropic\\.claude-[a-z]+-(?:4[-._](?:[89]|[1-9]\\d)(?!\\d)|(?:[5-9]|[1-9]\\d)(?!\\d)(?:[-._]\\d{1,2}(?!\\d))?)",
+                "description": "Bedrock Invoke ids for Claude 4.8 or higher: anthropic.claude-<family> with minor 4.8 through 4.99, any 5.x or later major-minor, or a bare 5+ major, which also admits new families such as fable. These models accept mid-conversation role system messages in place (verified live on Opus 4.8, Sonnet 5 and Fable 5), so unmapped future Bedrock Claudes keep the cache-preserving in-place handling instead of the hoist-all default. Listed first so bare-id provider inference, which takes the first pattern hit, resolves these Bedrock ids to bedrock; model-info resolution skips provider-mismatched rules either way.",
+                "extends": "anthropic-claude",
+                "model_info": {
+                    "litellm_provider": "bedrock",
+                    "supports_adaptive_thinking": true,
+                    "supports_mid_conversation_system": true
+                }
+            },
             {
                 "name": "anthropic-claude-adaptive-thinking",
                 "pattern": "(?:opus|sonnet|haiku)[-._](?:4[-._](?:[6-9]|[1-9]\\d)(?!\\d)|(?:[5-9]|[1-9]\\d{1,})[-._]\\d{1,2}(?!\\d))",