diff --git a/flow_judge/__init__.py b/flow_judge/__init__.py
index 6255276..e3989cb 100644
--- a/flow_judge/__init__.py
+++ b/flow_judge/__init__.py
@@ -3,7 +3,6 @@
 from flow_judge.eval_data_types import EvalInput, EvalOutput
 from flow_judge.flow_judge import AsyncFlowJudge, FlowJudge
 from flow_judge.metrics import CustomMetric, Metric, RubricItem, list_all_metrics
-from flow_judge.models import Hf, Llamafile, Vllm
 from flow_judge.models.common import BaseFlowJudgeModel
 from flow_judge.utils.prompt_formatter import format_rubric, format_user_prompt, format_vars
 
@@ -25,10 +24,39 @@
     "CustomMetric",
     "BaseFlowJudgeModel",
     "EvalOutput",
-    "Llamafile",
-    "Vllm",
-    "Hf",
 ]
 
+# Conditional imports for optional dependencies
+try:
+    from flow_judge.models.hf import Hf
+    __all__.append("Hf")
+except ImportError:
+    Hf = None
+
+try:
+    from flow_judge.models.vllm import Vllm
+    __all__.append("Vllm")
+except ImportError:
+    Vllm = None
+
+try:
+    from flow_judge.models.llamafile import Llamafile
+    __all__.append("Llamafile")
+except ImportError:
+    Llamafile = None
+
+def get_available_models():
+    """Return a list of available model classes based on installed extras."""
+    models = [BaseFlowJudgeModel]
+    if Hf is not None:
+        models.append(Hf)
+    if Vllm is not None:
+        models.append(Vllm)
+    if Llamafile is not None:
+        models.append(Llamafile)
+    return models
+
+__all__.append("get_available_models")
+
 # Add all metric names to __all__
 __all__ += list_all_metrics()
diff --git a/flow_judge/models/huggingface.py b/flow_judge/models/huggingface.py
index eb3461b..da0944e 100644
--- a/flow_judge/models/huggingface.py
+++ b/flow_judge/models/huggingface.py
@@ -4,6 +4,8 @@
 
 from pydantic import BaseModel
 
+from flow_judge.models.common import BaseFlowJudgeModel, ModelConfig, ModelType
+
 try:
     import torch
     from huggingface_hub import snapshot_download
@@ -15,8 +17,6 @@
 
 from tqdm import tqdm
 
-from flow_judge.models.common import BaseFlowJudgeModel, ModelConfig, ModelType
-
 logger = logging.getLogger(__name__)
 
 
diff --git a/flow_judge/models/llamafile.py b/flow_judge/models/llamafile.py
index 45e6021..f3de89e 100644
--- a/flow_judge/models/llamafile.py
+++ b/flow_judge/models/llamafile.py
@@ -19,7 +19,6 @@
 
 try:
     from openai import AsyncOpenAI, OpenAI
-
     LLAMAFILE_AVAILABLE = True
 except ImportError:
     LLAMAFILE_AVAILABLE = False
diff --git a/flow_judge/models/vllm.py b/flow_judge/models/vllm.py
index 627621f..a2af319 100644
--- a/flow_judge/models/vllm.py
+++ b/flow_judge/models/vllm.py
@@ -1,9 +1,6 @@
 import asyncio
 from typing import Any, Dict
 
-import torch
-from transformers import AutoTokenizer
-
 from flow_judge.models.common import (
     AsyncBaseFlowJudgeModel,
     BaseFlowJudgeModel,
@@ -12,8 +9,9 @@
 )
 
 try:
+    import torch
+    from transformers import AutoTokenizer
     from vllm import LLM, AsyncEngineArgs, AsyncLLMEngine, SamplingParams
-
     VLLM_AVAILABLE = True
 except ImportError:
     VLLM_AVAILABLE = False