diff --git a/app/services/openrouter.py b/app/services/openrouter.py
index c1ddb4f..ca31f2f 100644
--- a/app/services/openrouter.py
+++ b/app/services/openrouter.py
@@ -20,6 +20,31 @@ from app.config import get_settings
 OPENROUTER_URL = "https://openrouter.ai/api/v1/chat/completions"
 
 
+# Per-model USD rates: (input_per_million, output_per_million).
+# OpenRouter returns `usage.cost` directly; DeepSeek's native API does not.
+# Used as a fallback when the upstream omits the cost field.
+_MODEL_PRICING_USD_PER_MILLION: dict[str, tuple[float, float]] = {
+    "deepseek-v4-flash":          (0.07, 0.28),
+    "deepseek/deepseek-v4-flash": (0.07, 0.28),
+    "deepseek-chat":              (0.27, 1.10),
+    "deepseek-reasoner":          (0.55, 2.19),
+}
+
+
+def _estimate_cost_usd(model: str, prompt_tokens, completion_tokens) -> float | None:
+    """Compute cost from token counts when the upstream didn't return one.
+
+    Returns None if either token count is missing or the model isn't in
+    the pricing table — caller falls back to whatever value the upstream
+    did (or didn't) return.
+    """
+    rates = _MODEL_PRICING_USD_PER_MILLION.get(model)
+    if rates is None or prompt_tokens is None or completion_tokens is None:
+        return None
+    in_rate, out_rate = rates
+    return (prompt_tokens * in_rate + completion_tokens * out_rate) / 1_000_000.0
+
+
 @dataclass
 class LogResult:
     content: str
@@ -141,13 +166,21 @@ async def _call_provider(
             f"provider={provider}, model={used_model}, max_tokens={max_tokens})"
         )
     usage = data.get("usage") or {}
+    prompt_tokens = usage.get("prompt_tokens")
+    completion_tokens = usage.get("completion_tokens")
+    # OpenRouter populates `usage.cost`; DeepSeek's native API doesn't —
+    # estimate from tokens × per-model rates so the cost ledger stays
+    # populated regardless of which provider answered.
+    cost_usd = usage.get("cost") or usage.get("total_cost")
+    if cost_usd is None:
+        cost_usd = _estimate_cost_usd(used_model, prompt_tokens, completion_tokens)
     return LogResult(
         content=content,
         # Record provider+model so admin can see which path produced this row.
         model=f"{provider}/{used_model}",
-        prompt_tokens=usage.get("prompt_tokens"),
-        completion_tokens=usage.get("completion_tokens"),
-        cost_usd=usage.get("cost") or usage.get("total_cost"),
+        prompt_tokens=prompt_tokens,
+        completion_tokens=completion_tokens,
+        cost_usd=cost_usd,
     )