diff --git a/docs/gateway/clients.md b/docs/gateway/clients.md
index 0e8bc3538947..dbc85e097d63 100644
--- a/docs/gateway/clients.md
+++ b/docs/gateway/clients.md
@@ -127,10 +127,12 @@ if (attachments) {
 Both values are decoded per-attachment ceilings. Still check the serialized
 request against `policy.maxPayload`: attachments travel as base64, so a file near
 `maxBytes` can exceed the frame limit on its own. Older gateways omit
-`policy.attachments`; when it is absent, send and handle the server's typed
-rejection. Accepted MIME types and per-message counts are not advertised because
-they depend on the entrypoint and the resolved model, and the values are a
-connection-time snapshot, so re-read them on every reconnect.
+`policy.attachments`; when it is absent, send and handle the server outcome.
+Accepted MIME types and per-message handling are not advertised because they
+depend on the entrypoint and the resolved model. The gateway can return a typed
+rejection, while text-only model runs can omit additional images after their
+offload cap and still complete the request. The values are a connection-time
+snapshot, so re-read them on every reconnect.
 
 ## Recover state after reconnect
 
diff --git a/docs/gateway/protocol.md b/docs/gateway/protocol.md
index b43813de0259..14f440bcba45 100644
--- a/docs/gateway/protocol.md
+++ b/docs/gateway/protocol.md
@@ -181,9 +181,10 @@ Validating before send:
    promise the frame fits: attachments travel as base64, so a 20 MB file is about
    26.7 MB on the wire and exceeds the default 25 MiB frame limit on its own.
 3. Treat the server as authoritative for everything else. Accepted MIME types and
-   per-message counts are deliberately not advertised because they depend on the
-   entrypoint, the resolved model, and payload sniffing; the gateway still returns
-   a typed rejection.
+   per-message handling are deliberately not advertised because they depend on
+   the entrypoint, the resolved model, and payload sniffing. The gateway can
+   return a typed rejection, while text-only model runs can omit additional
+   images after their offload cap and still complete the request.
 4. Re-read the values on every reconnect. They are a connection-time snapshot, so
    a live `mediaMaxMb` edit reaches existing connections only after they reconnect.