diff --git a/packages/core/src/session/model-request.ts b/packages/core/src/session/model-request.ts
index 9a9292eb6beb..238a5b021c96 100644
--- a/packages/core/src/session/model-request.ts
+++ b/packages/core/src/session/model-request.ts
@@ -229,7 +229,7 @@ export const layer = Layer.effect(
       const entries = Object.entries(shaped.options)
       const generation = Object.fromEntries(entries.filter(([k]) => GENERATION_KEYS.has(k))) as GenerationOptionsFields
       const providerOptions = Object.fromEntries(entries.filter(([k]) => !GENERATION_KEYS.has(k)))
-      const root = session.fork?.sessionID ?? session.id
+      const affinity = session.parentID ?? session.fork?.sessionID ?? session.id
       const base = LLM.request({
         model: model.model,
         http: {
@@ -244,7 +244,7 @@ export const layer = Layer.effect(
           },
         },
         // TODO: Persist cache lineage so nested forks reuse the root session's cache key.
-        promptCacheKey: /^ses_[0-9a-f]{64}$/.test(root) ? root.slice(4) : root,
+        promptCacheKey: /^ses_[0-9a-f]{64}$/.test(affinity) ? affinity.slice(4) : affinity,
         system: shaped.system,
         messages: boundImages(unsupportedParts(shaped.messages, model.capabilities)),
         tools: Array.from(hooked, ([name, t]) => ({ ...t, name })),
diff --git a/packages/core/test/session-compaction.test.ts b/packages/core/test/session-compaction.test.ts
index e98bda151d5c..18a020b73726 100644
--- a/packages/core/test/session-compaction.test.ts
+++ b/packages/core/test/session-compaction.test.ts
@@ -397,7 +397,7 @@ it.effect("manual compaction summarizes short context instead of no-op", () =>
     ])
 
     expect(requests).toHaveLength(1)
-    expect(requests[0]?.promptCacheKey).toBe(sessionID)
+    expect(requests[0]?.promptCacheKey).toBe(parentID)
     expect(requests[0]?.http?.headers).toEqual({
       "x-session-affinity": sessionID,
       "X-Session-Id": sessionID,
diff --git a/packages/core/test/session-runner.test.ts b/packages/core/test/session-runner.test.ts
index 66f1af14af0d..f41f61e74725 100644
--- a/packages/core/test/session-runner.test.ts
+++ b/packages/core/test/session-runner.test.ts
@@ -4411,7 +4411,7 @@ describe("SessionRunnerLLM", () => {
     })
   })
 
-  scenario("adds the parent session header to child model requests", function* (s) {
+  scenario("uses parent cache affinity for child model requests", function* (s) {
     const parentID = Session.ID.make("ses_runner_parent")
 
     yield* s.db
@@ -4423,6 +4423,7 @@ describe("SessionRunnerLLM", () => {
     yield* s.runPrompt("Run child request")
 
     expect(s.requests[0]?.http?.headers?.["x-parent-session-id"]).toBe(parentID)
+    expect(s.requests[0]?.promptCacheKey).toBe(parentID)
   })
 
   scenario("runs different sessions concurrently", function* (s) {