diff --git a/app/agent/hooks/conductor_telemetry.py b/app/agent/hooks/conductor_telemetry.py
index 74d04c23..8e940760 100644
--- a/app/agent/hooks/conductor_telemetry.py
+++ b/app/agent/hooks/conductor_telemetry.py
@@ -125,6 +125,7 @@ async def wrap_model_call(
             tokens_in=_counter(usage.get("input")),
             tokens_out=_counter(usage.get("output")),
             cache_read_tokens=_counter(usage.get("cache")),
+            cache_write_tokens=_counter(usage.get("cache_write")),
             reasoning_tokens=_counter(usage.get("thoughts")),
             tool_use_tokens=_counter(usage.get("tool_use")),
             estimated_cost_usd_micros=_cost_micros(usage),
@@ -267,6 +268,7 @@ def _record(
         tokens_in: int = 0,
         tokens_out: int = 0,
         cache_read_tokens: int = 0,
+        cache_write_tokens: int = 0,
         reasoning_tokens: int = 0,
         tool_use_tokens: int = 0,
         tool_name: str | None = None,
@@ -303,6 +305,7 @@ def _record(
                 TelemetryField.TOKENS_IN: tokens_in,
                 TelemetryField.TOKENS_OUT: tokens_out,
                 TelemetryField.CACHE_READ_TOKENS: cache_read_tokens,
+                TelemetryField.CACHE_WRITE_TOKENS: cache_write_tokens,
                 TelemetryField.REASONING_TOKENS: reasoning_tokens,
                 TelemetryField.TOOL_USE_TOKENS: tool_use_tokens,
                 TelemetryField.TOOL_NAME: tool_name,
diff --git a/app/conductor/constants/telemetry.py b/app/conductor/constants/telemetry.py
index 1cdd9ec5..890962e8 100644
--- a/app/conductor/constants/telemetry.py
+++ b/app/conductor/constants/telemetry.py
@@ -76,6 +76,7 @@ class TelemetryField(StrEnum):
     MODEL = "model"
     RESPONSE_MODEL = "response_model"
     CACHE_READ_TOKENS = "cache_read_tokens"
+    CACHE_WRITE_TOKENS = "cache_write_tokens"
     REASONING_TOKENS = "reasoning_tokens"
     TOOL_USE_TOKENS = "tool_use_tokens"
     TOOL_NAME = "tool_name"
@@ -108,6 +109,7 @@ class TelemetryBatchField(StrEnum):
         TelemetryField.TOKENS_IN.value,
         TelemetryField.TOKENS_OUT.value,
         TelemetryField.CACHE_READ_TOKENS.value,
+        TelemetryField.CACHE_WRITE_TOKENS.value,
         TelemetryField.REASONING_TOKENS.value,
         TelemetryField.TOOL_USE_TOKENS.value,
     }
diff --git a/tests/conductor/test_telemetry.py b/tests/conductor/test_telemetry.py
index 509510e7..8aa63063 100644
--- a/tests/conductor/test_telemetry.py
+++ b/tests/conductor/test_telemetry.py
@@ -117,6 +117,7 @@ async def model_handler(_request: ModelRequest) -> AssistantMessage:
                     "input": 120,
                     "output": 40,
                     "cache": 20,
+                    "cache_write": 15,
                     "thoughts": 10,
                     "cost": {"estimated_usd": 0.00125},
                 },
@@ -143,6 +144,8 @@ async def tool_handler(_ctx, _state, _call) -> str:
     assert events[0][TelemetryField.MODEL] == "gpt-5.1"
     assert events[0][TelemetryField.TOKENS_IN] == 120
     assert events[0][TelemetryField.TOKENS_OUT] == 40
+    assert events[0][TelemetryField.CACHE_READ_TOKENS] == 20
+    assert events[0][TelemetryField.CACHE_WRITE_TOKENS] == 15
     assert events[0][TelemetryField.RESPONSE_MODEL] == "openai:gpt-5.1"
     assert events[0][TelemetryField.ESTIMATED_COST_USD_MICROS] == 1250
     assert events[0][TelemetryField.COST_SOURCE] == "evoflux_catalog"