diff --git a/litellm/utils.py b/litellm/utils.py
index 8ef55758ef1..c0aef30bebd 100644
--- a/litellm/utils.py
+++ b/litellm/utils.py
@@ -3523,6 +3523,9 @@ def get_optional_params_embeddings(
     special_params: Final = passed_params.pop("kwargs")
 
     drop_params = normalize_drop_params(passed_params.pop("drop_params", None))
+    resolved_drop_params = drop_params
+    if resolved_drop_params is None:
+        resolved_drop_params = normalize_drop_params(litellm.drop_params)
     additional_drop_params = passed_params.pop("additional_drop_params", None)
     allowed_openai_params = passed_params.pop("allowed_openai_params", None) or []
     # Remove function objects from passed_params to avoid JSON serialization errors
@@ -3568,7 +3571,7 @@ def get_optional_params_embeddings(
             non_default_params=non_default_params,
             optional_params={},
             model=model,
-            drop_params=drop_params if drop_params is not None else False,
+            drop_params=resolved_drop_params if resolved_drop_params is not None else False,
         )
         # Provider-only params (e.g. Cohere input_type) are not in
         # OPENAI_EMBEDDING_PARAMS, so embedding_pre_process drops them from
@@ -3842,6 +3845,14 @@ def get_optional_params_embeddings(
             optional_params = non_default_params
     else:
         optional_params = non_default_params
+        if (
+            resolved_drop_params is True
+            and (custom_llm_provider == "azure" or custom_llm_provider in litellm.openai_compatible_providers)
+            and "text-embedding-3" not in model
+            and "dimensions" in optional_params
+            and "dimensions" not in (allowed_openai_params or ())
+        ):
+            optional_params.pop("dimensions", None)
 
     final_params = add_provider_specific_params_to_optional_params(
         optional_params=optional_params,
diff --git a/tests/test_litellm/litellm_core_utils/test_embedding_optional_params.py b/tests/test_litellm/litellm_core_utils/test_embedding_optional_params.py
new file mode 100644
index 00000000000..376467685be
--- /dev/null
+++ b/tests/test_litellm/litellm_core_utils/test_embedding_optional_params.py
@@ -0,0 +1,67 @@
+import pytest
+
+import litellm
+
+
+@pytest.mark.parametrize("provider", ["azure", "together_ai"])
+def test_embedding_dimensions_drop_params_for_openai_compatible_provider(provider, monkeypatch):
+    monkeypatch.setattr(litellm, "drop_params", False)
+    dropped = litellm.utils.get_optional_params_embeddings(
+        model=f"{provider}/dummy-model",
+        custom_llm_provider=provider,
+        dimensions=512,
+        drop_params=True,
+    )
+    assert "dimensions" not in dropped
+
+    monkeypatch.setattr(litellm, "drop_params", True)
+    dropped_globally = litellm.utils.get_optional_params_embeddings(
+        model=f"{provider}/dummy-model",
+        custom_llm_provider=provider,
+        dimensions=512,
+    )
+    assert "dimensions" not in dropped_globally
+
+    monkeypatch.setattr(litellm, "drop_params", False)
+    preserved = litellm.utils.get_optional_params_embeddings(
+        model=f"{provider}/dummy-model",
+        custom_llm_provider=provider,
+        dimensions=512,
+    )
+    assert preserved["dimensions"] == 512
+
+    monkeypatch.setattr(litellm, "drop_params", True)
+    model_supported = litellm.utils.get_optional_params_embeddings(
+        model=f"{provider}/text-embedding-3-small",
+        custom_llm_provider=provider,
+        dimensions=512,
+    )
+    assert model_supported["dimensions"] == 512
+
+    explicitly_allowed = litellm.utils.get_optional_params_embeddings(
+        model=f"{provider}/legacy-model",
+        custom_llm_provider=provider,
+        dimensions=512,
+        allowed_openai_params=["dimensions"],
+    )
+    assert explicitly_allowed["dimensions"] == 512
+
+
+@pytest.mark.parametrize(
+    ("provider", "model"),
+    [
+        ("nvidia_nim", "nvidia_nim/nv-embedqa-e5-v5"),
+        ("fireworks_ai", "fireworks_ai/nomic-ai/nomic-embed-text-v1.5"),
+        ("dashscope", "dashscope/text-embedding-v3"),
+        ("hosted_vllm", "hosted_vllm/Qwen/Qwen3-Embedding-0.6B"),
+    ],
+)
+def test_embedding_dimensions_preserved_for_provider_mappings(provider, model):
+    optional_params = litellm.utils.get_optional_params_embeddings(
+        model=model,
+        custom_llm_provider=provider,
+        dimensions=128,
+        drop_params=True,
+    )
+
+    assert optional_params["dimensions"] == 128