diff --git a/backend/app/models/cost.py b/backend/app/models/cost.py
new file mode 100644
index 0000000..7550274
--- /dev/null
+++ b/backend/app/models/cost.py
@@ -0,0 +1,42 @@
+"""Cost tracking model - per-call usage logs."""
+from sqlalchemy import Column, String, Integer, Text, ForeignKey, Float
+from app.models.base import Base, TimestampMixin, TenantMixin
+
+
+class CostLog(Base, TimestampMixin, TenantMixin):
+    """Records every LLM call's token usage and cost.
+
+    Enables multi-dimensional cost analysis:
+      tenant x project x task x worker x model
+    """
+    __tablename__ = "cost_logs"
+
+    id = Column(Integer, primary_key=True, autoincrement=True)
+
+    project_id = Column(Integer, ForeignKey("projects.id"), nullable=True, index=True)
+    task_id = Column(Integer, ForeignKey("tasks.id"), nullable=True, index=True)
+    worker_id = Column(Integer, ForeignKey("ai_workers.id"), nullable=True, index=True)
+
+    # Model info
+    provider = Column(String(100), nullable=False)
+    model_name = Column(String(200), nullable=False)
+
+    # Token counts
+    prompt_tokens = Column(Integer, default=0)
+    completion_tokens = Column(Integer, default=0)
+    total_tokens = Column(Integer, default=0)
+
+    # Cost in cents
+    cost_cents = Column(Integer, default=0)
+
+    # Duration
+    duration_ms = Column(Integer, default=0)
+
+    # Status
+    status = Column(String(50), default="success")  # success, error, timeout
+
+    # Error message if any
+    error_message = Column(Text, nullable=True)
+
+    def __repr__(self):
+        return f"<CostLog {self.provider}/{self.model_name} tokens={self.total_tokens}>"