diff --git a/docs/my-website/docs/proxy/virtual_keys.md b/docs/my-website/docs/proxy/virtual_keys.md
index 2be4b95c1f4..83994701c49 100644
--- a/docs/my-website/docs/proxy/virtual_keys.md
+++ b/docs/my-website/docs/proxy/virtual_keys.md
@@ -93,6 +93,7 @@ Request Params:
 - `config`: *Optional[dict]* - any key-specific configs, overrides config in config.yaml
 - `spend`: *Optional[int]* - Amount spent by key. Default is 0. Will be updated by proxy whenever key is used. https://docs.litellm.ai/docs/proxy/virtual_keys#managing-auth---tracking-spend
 - `max_budget`: *Optional[float]* - Specify max budget for a given key.
+- `model_max_budget`: *Optional[dict[str, float]]* - Specify max budget for each model, `model_max_budget={"gpt4": 0.5, "gpt-5": 0.01}`
 - `max_parallel_requests`: *Optional[int]* - Rate limit a user based on the number of parallel requests. Raises 429 error, if user's parallel requests > x.
 - `metadata`: *Optional[dict]* - Metadata for key, store information for key. Example metadata = {"team": "core-infra", "app": "app2", "email": "ishaan@berri.ai" }
 
@@ -676,8 +677,6 @@ general_settings:
 
 ### [BETA] Dynamo DB 
 
-Only live in `v1.16.21.dev1`. 
-
 #### Step 1. Save keys to env
 
 ```shell
diff --git a/litellm/proxy/_types.py b/litellm/proxy/_types.py
index 372b953e08d..5d74b9dedeb 100644
--- a/litellm/proxy/_types.py
+++ b/litellm/proxy/_types.py
@@ -155,6 +155,9 @@ class GenerateKeyRequest(GenerateRequestBase):
     aliases: Optional[dict] = {}
     config: Optional[dict] = {}
     permissions: Optional[dict] = {}
+    model_max_budget: Optional[dict] = (
+        {}
+    )  # {"gpt-4": 5.0, "gpt-3.5-turbo": 5.0}, defaults to {}
 
 
 class GenerateKeyResponse(GenerateKeyRequest):
@@ -167,7 +170,13 @@ class GenerateKeyResponse(GenerateKeyRequest):
     def set_model_info(cls, values):
         if values.get("token") is not None:
             values.update({"key": values.get("token")})
-        dict_fields = ["metadata", "aliases", "config", "permissions"]
+        dict_fields = [
+            "metadata",
+            "aliases",
+            "config",
+            "permissions",
+            "model_max_budget",
+        ]
         for field in dict_fields:
             value = values.get(field)
             if value is not None and isinstance(value, str):
@@ -383,6 +392,8 @@ class LiteLLM_VerificationToken(LiteLLMBase):
     budget_reset_at: Optional[datetime] = None
     allowed_cache_controls: Optional[list] = []
     permissions: Dict = {}
+    model_spend: Dict = {}
+    model_max_budget: Dict = {}
 
 
 class UserAPIKeyAuth(
@@ -410,6 +421,8 @@ class LiteLLM_UserTable(LiteLLMBase):
     user_id: str
     max_budget: Optional[float]
     spend: float = 0.0
+    model_max_budget: Optional[Dict] = {}
+    model_spend: Optional[Dict] = {}
     user_email: Optional[str]
     models: list = []
 
diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py
index a02042ab592..bd30a621fe1 100644
--- a/litellm/proxy/proxy_server.py
+++ b/litellm/proxy/proxy_server.py
@@ -940,13 +940,26 @@ async def update_database(
                     # Calculate the new cost by adding the existing cost and response_cost
                     existing_spend_obj.spend = existing_spend + response_cost
 
+                    # track cost per model, for the given user
+                    spend_per_model = existing_spend_obj.model_spend or {}
+                    current_model = kwargs.get("model")
+
+                    if current_model is not None and spend_per_model is not None:
+                        if spend_per_model.get(current_model) is None:
+                            spend_per_model[current_model] = response_cost
+                        else:
+                            spend_per_model[current_model] += response_cost
+                    existing_spend_obj.model_spend = spend_per_model
+
                     valid_token = user_api_key_cache.get_cache(key=id)
                     if valid_token is not None and isinstance(valid_token, dict):
                         user_api_key_cache.set_cache(
                             key=id, value=existing_spend_obj.json()
                         )
 
-                    verbose_proxy_logger.debug(f"new cost: {existing_spend_obj.spend}")
+                    verbose_proxy_logger.debug(
+                        f"user - new cost: {existing_spend_obj.spend}, user_id: {id}"
+                    )
                     data_list.append(existing_spend_obj)
 
                     # Update the cost column for the given user id
@@ -983,15 +996,28 @@ async def update_database(
                     # Calculate the new cost by adding the existing cost and response_cost
                     new_spend = existing_spend + response_cost
 
-                    verbose_proxy_logger.debug(f"new cost: {new_spend}")
+                    # track cost per model, for the given key
+                    spend_per_model = existing_spend_obj.model_spend or {}
+                    current_model = kwargs.get("model")
+                    if current_model is not None and spend_per_model is not None:
+                        if spend_per_model.get(current_model) is None:
+                            spend_per_model[current_model] = response_cost
+                        else:
+                            spend_per_model[current_model] += response_cost
+
+                    verbose_proxy_logger.debug(
+                        f"new cost: {new_spend}, new spend per model: {spend_per_model}"
+                    )
                     # Update the cost column for the given token
                     await prisma_client.update_data(
-                        token=token, data={"spend": new_spend}
+                        token=token,
+                        data={"spend": new_spend, "model_spend": spend_per_model},
                     )
 
                     valid_token = user_api_key_cache.get_cache(key=token)
                     if valid_token is not None:
                         valid_token.spend = new_spend
+                        valid_token.model_spend = spend_per_model
                         user_api_key_cache.set_cache(key=token, value=valid_token)
                 elif custom_db_client is not None:
                     # Fetch the existing cost for the given token
@@ -1071,10 +1097,21 @@ async def update_database(
                     # Calculate the new cost by adding the existing cost and response_cost
                     new_spend = existing_spend + response_cost
 
+                    # track cost per model, for the given team
+                    spend_per_model = existing_spend_obj.model_spend or {}
+                    current_model = kwargs.get("model")
+                    if current_model is not None and spend_per_model is not None:
+                        if spend_per_model.get(current_model) is None:
+                            spend_per_model[current_model] = response_cost
+                        else:
+                            spend_per_model[current_model] += response_cost
+
                     verbose_proxy_logger.debug(f"new cost: {new_spend}")
                     # Update the cost column for the given token
                     await prisma_client.update_data(
-                        team_id=team_id, data={"spend": new_spend}, table_name="team"
+                        team_id=team_id,
+                        data={"spend": new_spend, "model_spend": spend_per_model},
+                        table_name="team",
                     )
 
                 elif custom_db_client is not None:
@@ -1648,6 +1685,7 @@ async def generate_key_helper_fn(
     key_alias: Optional[str] = None,
     allowed_cache_controls: Optional[list] = [],
     permissions: Optional[dict] = {},
+    model_max_budget: Optional[dict] = {},
 ):
     global prisma_client, custom_db_client, user_api_key_cache
 
@@ -1681,6 +1719,8 @@ async def generate_key_helper_fn(
     config_json = json.dumps(config)
     permissions_json = json.dumps(permissions)
     metadata_json = json.dumps(metadata)
+    model_max_budget_json = json.dumps(model_max_budget)
+
     user_id = user_id or str(uuid.uuid4())
     user_role = user_role or "app_user"
     tpm_limit = tpm_limit
@@ -1723,6 +1763,7 @@ async def generate_key_helper_fn(
             "budget_reset_at": key_reset_at,
             "allowed_cache_controls": allowed_cache_controls,
             "permissions": permissions_json,
+            "model_max_budget": model_max_budget_json,
         }
         if (
             general_settings.get("allow_user_auth", False) == True
@@ -1738,6 +1779,11 @@ async def generate_key_helper_fn(
             saved_token["metadata"] = json.loads(saved_token["metadata"])
         if isinstance(saved_token["permissions"], str):
             saved_token["permissions"] = json.loads(saved_token["permissions"])
+        if isinstance(saved_token["model_max_budget"], str):
+            saved_token["model_max_budget"] = json.loads(
+                saved_token["model_max_budget"]
+            )
+
         if saved_token.get("expires", None) is not None and isinstance(
             saved_token["expires"], datetime
         ):
@@ -3081,6 +3127,7 @@ async def generate_key_fn(
     - max_parallel_requests: Optional[int] - Rate limit a user based on the number of parallel requests. Raises 429 error, if user's parallel requests > x.
     - metadata: Optional[dict] - Metadata for key, store information for key. Example metadata = {"team": "core-infra", "app": "app2", "email": "ishaan@berri.ai" }
     - permissions: Optional[dict] - key-specific permissions. Currently just used for turning off pii masking (if connected). Example - {"pii": false}
+    - model_max_budget: Optional[dict] - key-specific model budget in USD. Example - {"text-davinci-002": 0.5, "gpt-3.5-turbo": 0.5}. IF null or {} then no model specific budget.
 
     Returns:
     - key: (str) The generated api key
diff --git a/litellm/proxy/schema.prisma b/litellm/proxy/schema.prisma
index 5a57b880842..101cf9b7f09 100644
--- a/litellm/proxy/schema.prisma
+++ b/litellm/proxy/schema.prisma
@@ -24,6 +24,8 @@ model LiteLLM_TeamTable {
     budget_reset_at DateTime?
     created_at    DateTime               @default(now()) @map("created_at")
     updated_at    DateTime               @default(now()) @updatedAt @map("updated_at")
+    model_spend      Json @default("{}")
+    model_max_budget Json @default("{}")
 }
 
 // Track spend, rate limit, budget Users
@@ -41,6 +43,8 @@ model LiteLLM_UserTable {
     budget_duration String? 
     budget_reset_at DateTime?
     allowed_cache_controls String[] @default([])
+    model_spend      Json @default("{}")
+    model_max_budget Json @default("{}")
 }
 
 // Generate Tokens for Proxy
@@ -64,6 +68,8 @@ model LiteLLM_VerificationToken {
     budget_duration String? 
     budget_reset_at DateTime?
     allowed_cache_controls String[] @default([])
+    model_spend      Json @default("{}")
+    model_max_budget Json @default("{}")
 }
 
 // store proxy config.yaml
diff --git a/schema.prisma b/schema.prisma
index 5a57b880842..101cf9b7f09 100644
--- a/schema.prisma
+++ b/schema.prisma
@@ -24,6 +24,8 @@ model LiteLLM_TeamTable {
     budget_reset_at DateTime?
     created_at    DateTime               @default(now()) @map("created_at")
     updated_at    DateTime               @default(now()) @updatedAt @map("updated_at")
+    model_spend      Json @default("{}")
+    model_max_budget Json @default("{}")
 }
 
 // Track spend, rate limit, budget Users
@@ -41,6 +43,8 @@ model LiteLLM_UserTable {
     budget_duration String? 
     budget_reset_at DateTime?
     allowed_cache_controls String[] @default([])
+    model_spend      Json @default("{}")
+    model_max_budget Json @default("{}")
 }
 
 // Generate Tokens for Proxy
@@ -64,6 +68,8 @@ model LiteLLM_VerificationToken {
     budget_duration String? 
     budget_reset_at DateTime?
     allowed_cache_controls String[] @default([])
+    model_spend      Json @default("{}")
+    model_max_budget Json @default("{}")
 }
 
 // store proxy config.yaml