diff --git a/renderers/client.py b/renderers/client.py
index 9bf008050..930ad0c89 100644
--- a/renderers/client.py
+++ b/renderers/client.py
@@ -261,9 +261,11 @@ async def generate(
     Returns a dict with: request_id, prompt_ids, renderer_prompt_ids,
     mm_placeholders, completion_ids, completion_logprobs, content,
     reasoning_content, tool_calls, finish_reason, routed_experts,
-    multi_modal_data, prompt_attribution. ``renderer_prompt_ids`` is the
-    unexpanded logical prompt when ``process_multimodal=False`` and ``None``
-    otherwise.
+    weight_versions, multi_modal_data, prompt_attribution.
+    ``renderer_prompt_ids`` is the unexpanded logical prompt when
+    ``process_multimodal=False`` and ``None`` otherwise. ``weight_versions``
+    is vLLM's (>= 0.32) list of ``{version, start, end}`` spans over
+    completion-token indexes, or ``None`` when the engine labels none.
 
     ``prompt_attribution`` is the renderer's :class:`RenderedTokens` for
     the prompt — either the one this call computed via
@@ -431,6 +433,7 @@ def _prepare():
         "reasoning_complete": parsed.reasoning_complete,
         "routed_experts": routed_experts,
         "sampling_mask": sampling_mask,
+        "weight_versions": choice.get("weight_versions"),
         # The mm sidecar consumed on the request side, surfaced back so
         # callers can persist it on the trajectory step for downstream
         # multi-turn bridging and training-sample construction.