diff --git a/CHANGELOG.md b/CHANGELOG.md
index 555a8f6..b1eed10 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,7 +2,7 @@
 
 All notable changes to this project will be documented here. Format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); versions follow [SemVer](https://semver.org).
 
-## [Unreleased]
+## [0.4.0] - 2026-09-07
 
 ### Added
 
diff --git a/pyproject.toml b/pyproject.toml
index 3805cbe..8492734 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "lago-agent-sdk"
-version = "0.3.0"
+version = "0.4.0"
 description = "Instrument LLM clients and emit usage events to Lago."
 readme = "README.md"
 requires-python = ">=3.10"
diff --git a/src/lago_agent_sdk/__init__.py b/src/lago_agent_sdk/__init__.py
index c3548e7..8afe314 100644
--- a/src/lago_agent_sdk/__init__.py
+++ b/src/lago_agent_sdk/__init__.py
@@ -35,4 +35,4 @@
     "compute_cost",
     "TOKEN_BILLED_PROVIDERS",
 ]
-__version__ = "0.2.0"
+__version__ = "0.4.0"
diff --git a/src/lago_agent_sdk/token_semantics.py b/src/lago_agent_sdk/token_semantics.py
index 97f9ea7..acae9b7 100644
--- a/src/lago_agent_sdk/token_semantics.py
+++ b/src/lago_agent_sdk/token_semantics.py
@@ -153,10 +153,10 @@
         # reported input_tokens unchanged (18825) with cached_tokens 18810 INSIDE it,
         # total = input + output exactly (06b_real_cache_control_warm.json); reasoning
         # came back inside output (128 of 169, total = input + output,
-        # 07_real_reasoning.json). cache_write is inferred from the same normalization
-        # rather than measured — every observed write reported cache_write_tokens: 0
-        # while the warm read proved the cache existed, and the arithmetic keeps the
-        # write inside input — re-verify the day a nonzero write appears in a capture.
+        # 07_real_reasoning.json). cache_write follows the same convention, confirmed
+        # by measurement on 2026-09-07: an OpenAI-served cold write reported
+        # input_tokens_details.cache_write_tokens 4490 INSIDE input_tokens 4493, with
+        # total = input + output — so the write stays inside input here too.
         #
         # Router's OTHER surface, `/v1/messages` (stamped "ramp_router_messages" by the
         # Anthropic adapter), is deliberately ABSENT. It keeps Anthropic's additive
diff --git a/uv.lock b/uv.lock
index ac9ffde..f13d31c 100644
--- a/uv.lock
+++ b/uv.lock
@@ -784,7 +784,7 @@ wheels = [
 
 [[package]]
 name = "lago-agent-sdk"
-version = "0.3.0"
+version = "0.4.0"
 source = { editable = "." }
 dependencies = [
     { name = "requests" },