fix fireworks model name lookup and bedrock snake-case output_config leak

- Fireworks: add accounts/fireworks/models/ candidate key to
  _get_model_cost_capability so short-form model names (e.g.
  fireworks_ai/qwen3-8b) resolve to their full pricing JSON entry
- Bedrock: always pop snake-case output_config from inference_params to
  prevent it from leaking to the Converse API (which expects camelCase
  outputConfig); nullify the camelCase value when model doesn't support it

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

litellm/llms/bedrock/chat/converse_transformation.py

@@ -1203,12 +1203,13 @@ class AmazonConverseConfig(BaseConfig):
         output_config: Optional[OutputConfigBlock] = inference_params.pop(
             "outputConfig", None
         )
+        inference_params.pop("output_config", None)
         if not _supports_factory(
             model=strip_bedrock_routing_prefix(model),
             custom_llm_provider="bedrock",
             key="supports_output_config",
         ):
-            inference_params.pop("output_config", None)
+            output_config = None
 
         # keep supported params in 'inference_params', and set all model-specific params in 'additional_request_params'
         additional_request_params = {

litellm/llms/fireworks_ai/chat/transformation.py

@@ -254,6 +254,10 @@ class FireworksAIConfig(OpenAIGPTConfig):
         if not model.startswith("fireworks_ai/"):
             candidate_keys.append(f"fireworks_ai/{model}")
 
+        stripped = model.removeprefix("fireworks_ai/")
+        if not stripped.startswith("accounts/"):
+            candidate_keys.append(f"fireworks_ai/accounts/fireworks/models/{stripped}")
+
         for candidate_key in candidate_keys:
             model_info = litellm.model_cost.get(candidate_key)
             if model_info is not None and model_info.get(capability) is not None: