feat: add native Groq model integration for high-speed evaluations

deepeval/config/settings.py

@@ -250,9 +250,7 @@ def __setattr__(self, name: str, value):
         return super().__setattr__(name, value)
 
     model_config = SettingsConfigDict(
-        extra="ignore",
-        case_sensitive=True,
-        validate_assignment=True,
+        extra="ignore", case_sensitive=True, validate_assignment=True,
     )
 
     #
@@ -272,8 +270,7 @@ def __setattr__(self, name: str, value):
         description="Extra PYTHONPATH used by the CLI runner (default: current project '.').",
     )
     CONFIDENT_REGION: Optional[str] = Field(
-        None,
-        description="Optional Confident AI region hint (uppercased).",
+        None, description="Optional Confident AI region hint (uppercased).",
     )
     CONFIDENT_OPEN_BROWSER: Optional[bool] = Field(
         True,
@@ -361,8 +358,7 @@ def __setattr__(self, name: str, value):
         description="CUDA debug toggle (forces synchronous CUDA ops). Useful for debugging GPU errors.",
     )
     CUDA_VISIBLE_DEVICES: Optional[str] = Field(
-        None,
-        description="CUDA device visibility mask (e.g. '0' or '0,1').",
+        None, description="CUDA device visibility mask (e.g. '0' or '0,1').",
     )
     TOKENIZERS_PARALLELISM: Optional[bool] = Field(
         None,
@@ -529,6 +525,19 @@ def __setattr__(self, name: str, value):
     GROK_COST_PER_OUTPUT_TOKEN: Optional[float] = Field(
         None, description="Grok output token cost (used for cost reporting)."
     )
+    # Groq
+    USE_GROQ_MODEL: Optional[bool] = Field(
+        None, description="Select Groq as the active LLM provider."
+    )
+    GROQ_API_KEY: Optional[SecretStr] = Field(None, description="Groq API key.")
+    GROQ_MODEL_NAME: Optional[str] = Field(None, description="Groq model name.")
+    GROQ_COST_PER_INPUT_TOKEN: Optional[float] = Field(
+        None, description="Groq input token cost (used for cost reporting)."
+    )
+    GROQ_COST_PER_OUTPUT_TOKEN: Optional[float] = Field(
+        None, description="Groq output token cost (used for cost reporting)."
+    )
+
     # LiteLLM
     USE_LITELLM: Optional[bool] = Field(
         None, description="Select LiteLLM as the active LLM provider."
@@ -996,6 +1005,7 @@ def DEEPEVAL_TASK_GATHER_BUFFER_SECONDS(self) -> float:
         "USE_GEMINI_MODEL",
         "USE_MOONSHOT_MODEL",
         "USE_GROK_MODEL",
+        "USE_GROQ_MODEL",
         "USE_DEEPSEEK_MODEL",
         "USE_LITELLM",
         "USE_AZURE_OPENAI_EMBEDDING",
@@ -1396,9 +1406,7 @@ def __exit__(self, exc_type, exc, tb):
                 updates = {k: after[k] for k in changed_keys}
 
                 if "LOG_LEVEL" in updates:
-                    from deepeval.config.logging import (
-                        apply_deepeval_log_level,
-                    )
+                    from deepeval.config.logging import apply_deepeval_log_level
 
                     apply_deepeval_log_level()
 

deepeval/constants.py

@@ -36,6 +36,7 @@ class ProviderSlug(str, Enum):
     LOCAL = "local"
     OLLAMA = "ollama"
     OPENROUTER = "openrouter"
+    GROQ = "groq"
 
 
 def slugify(value: Union[str, ProviderSlug]) -> str:

deepeval/models/__init__.py

@@ -14,6 +14,7 @@
     LiteLLMModel,
     KimiModel,
     GrokModel,
+    GroqModel,
     DeepSeekModel,
     PortkeyModel,
     OpenRouterModel,
@@ -39,6 +40,7 @@
     "LiteLLMModel",
     "KimiModel",
     "GrokModel",
+    "GroqModel",
     "DeepSeekModel",
     "OpenAIEmbeddingModel",
     "AzureOpenAIEmbeddingModel",

deepeval/models/llms/__init__.py

@@ -11,6 +11,7 @@
 from .deepseek_model import DeepSeekModel
 from .portkey_model import PortkeyModel
 from .openrouter_model import OpenRouterModel
+from .groq_model import GroqModel
 
 __all__ = [
     "AzureOpenAIModel",
@@ -26,4 +27,5 @@
     "DeepSeekModel",
     "PortkeyModel",
     "OpenRouterModel",
+    "GroqModel",
 ]

deepeval/models/llms/constants.py

@@ -2,6 +2,7 @@
 
 from deepeval.models.base_model import DeepEvalModelData
 
+
 DEFAULT_GPT_MODEL = "gpt-5.4"
 # OpenRouter uses provider/model format (e.g., "openai/gpt-4", "anthropic/claude-3-opus")
 # DeepEval does not validate OpenRouter model strings.
@@ -2172,3 +2173,32 @@ def make_model_data(**kwargs: Any) -> ModelDataFactory:
         ),
     }
 )
+
+GROQ_MODELS_DATA = ModelDataRegistry(
+    {
+        "llama3-8b-8192": make_model_data(
+            supports_log_probs=False,
+            supports_multimodal=False,
+            supports_structured_outputs=True,
+            supports_json=True,
+            input_price=0.05 / 1e6,
+            output_price=0.08 / 1e6,
+        ),
+        "llama3-70b-8192": make_model_data(
+            supports_log_probs=False,
+            supports_multimodal=False,
+            supports_structured_outputs=True,
+            supports_json=True,
+            input_price=0.59 / 1e6,
+            output_price=0.79 / 1e6,
+        ),
+        "mixtral-8x7b-32768": make_model_data(
+            supports_log_probs=False,
+            supports_multimodal=False,
+            supports_structured_outputs=True,
+            supports_json=True,
+            input_price=0.24 / 1e6,
+            output_price=0.24 / 1e6,
+        ),
+    }
+)