diff --git a/apps/server/src/mcp/OrchestratorMcpService.activity.test.ts b/apps/server/src/mcp/OrchestratorMcpService.activity.test.ts
index c35b605dca4fb..dfaebe9228d52 100644
--- a/apps/server/src/mcp/OrchestratorMcpService.activity.test.ts
+++ b/apps/server/src/mcp/OrchestratorMcpService.activity.test.ts
@@ -136,6 +136,7 @@ it("readThread preserves caller access errors and prefers activity-run status ov
       makeRun({ id: cancelledRunId, ordinal: 2, status: "cancelled" }),
     ],
     visibleTurnItems: [],
+    providerTurns: [],
     runtimeRequests: [],
     messages: [],
     contextTransfers: [],
@@ -222,6 +223,7 @@ it("readThread pages cut-off text without splitting an emoji across pages", asyn
     }),
     runs: [makeRun({ id: activeRunId, ordinal: 1, status: "running" })],
     visibleTurnItems: [],
+    providerTurns: [],
     runtimeRequests: [],
     messages: [],
     contextTransfers: [],
@@ -312,6 +314,7 @@ it("taskStatus returns task.providerInstanceId rather than the driver kind", asy
     }),
     runs: [makeRun({ id: activeRunId, ordinal: 1, status: "running" })],
     visibleTurnItems: [],
+    providerTurns: [],
     runtimeRequests: [],
     messages: [],
     contextTransfers: [],
@@ -365,6 +368,7 @@ it("taskStatus returns task.providerInstanceId rather than the driver kind", asy
       }),
     ],
     visibleTurnItems: [],
+    providerTurns: [],
     runtimeRequests: [],
     messages: [],
     contextTransfers: [
@@ -438,6 +442,7 @@ it("readThread and sendToThread reach threads in other projects", async () => {
       return parentRuns;
     },
     visibleTurnItems: [],
+    providerTurns: [],
     runtimeRequests: [],
     messages: [],
     contextTransfers: [],
@@ -474,6 +479,7 @@ it("readThread and sendToThread reach threads in other projects", async () => {
           },
         },
       ],
+      providerTurns: [],
       runtimeRequests: [],
       messages: [],
       contextTransfers: [],
@@ -670,6 +676,7 @@ it("requestSecret saves a pasted value to a file without opening a card", async
     }),
     runs: [{ ...makeRun({ id: activeRunId, ordinal: 1, status: "running" }), rootNodeId: taskId }],
     visibleTurnItems: [],
+    providerTurns: [],
     runtimeRequests: [],
     messages: [],
     contextTransfers: [],
@@ -791,6 +798,7 @@ it("requestSecret returns the unsaved note for a declined, cancelled or timed-ou
           card === undefined
             ? []
             : [{ type: "secret_request", id: card.turnItemId, secretStatus: outcome }],
+        providerTurns: [],
         runtimeRequests: [],
         messages: [],
         contextTransfers: [],
diff --git a/apps/server/src/mcp/OrchestratorMcpService.test.ts b/apps/server/src/mcp/OrchestratorMcpService.test.ts
index f52ac95174277..df13d037005a4 100644
--- a/apps/server/src/mcp/OrchestratorMcpService.test.ts
+++ b/apps/server/src/mcp/OrchestratorMcpService.test.ts
@@ -145,6 +145,7 @@ describe("OrchestratorMcpService", () => {
         turnItems: [],
         subagents: [],
         providerThreads: [],
+        providerTurns: [],
         runtimeRequests: [],
       } as unknown as OrchestrationV2ThreadProjection;
       let hasNestedWork = true;
@@ -288,6 +289,7 @@ describe("OrchestratorMcpService", () => {
         messages: [],
         subagents: [],
         providerThreads: [],
+        providerTurns: [],
         runtimeRequests: [],
         turnItems: [],
       } as unknown as OrchestrationV2ThreadProjection;
@@ -385,6 +387,7 @@ describe("OrchestratorMcpService", () => {
           messages: [],
           subagents: [],
           providerThreads: [],
+          providerTurns: [],
           runtimeRequests: pendingRequest ? [{ status: "pending" }] : [],
           nodes: [],
           turnItems: [],
@@ -485,6 +488,7 @@ describe("OrchestratorMcpService", () => {
         messages: [],
         subagents: [],
         providerThreads: [],
+        providerTurns: [],
         runtimeRequests: [],
       } as unknown as OrchestrationV2ThreadProjection;
       const layerDependencies = Layer.mergeAll(
@@ -575,6 +579,7 @@ describe("OrchestratorMcpService", () => {
           messages: [],
           subagents: [],
           providerThreads: [],
+          providerTurns: [],
           runtimeRequests: [],
           turnItems: [],
         } as unknown as OrchestrationV2ThreadProjection;
@@ -678,6 +683,7 @@ describe("OrchestratorMcpService", () => {
         messages: [],
         subagents: [],
         providerThreads: [],
+        providerTurns: [],
         runtimeRequests: [],
       } as unknown as OrchestrationV2ThreadProjection;
       const layerDependencies = Layer.mergeAll(
@@ -761,6 +767,7 @@ describe("OrchestratorMcpService", () => {
         messages: [],
         subagents: [],
         providerThreads: [],
+        providerTurns: [],
         runtimeRequests: [],
       } as unknown as OrchestrationV2ThreadProjection;
       const layerDependencies = Layer.mergeAll(
@@ -850,6 +857,7 @@ describe("OrchestratorMcpService", () => {
         messages: [],
         subagents: [],
         providerThreads: [],
+        providerTurns: [],
         runtimeRequests: [],
       } as unknown as OrchestrationV2ThreadProjection;
       const layerDependencies = Layer.mergeAll(
@@ -973,6 +981,7 @@ describe("OrchestratorMcpService", () => {
               appOwnedTask("node:mcp-cancel-grandchild-below", childThreadId, grandchildThreadId),
             ],
             providerThreads: [],
+            providerTurns: [],
             runtimeRequests: [],
           },
         ],
@@ -997,6 +1006,7 @@ describe("OrchestratorMcpService", () => {
             messages: [],
             subagents: [],
             providerThreads: [],
+            providerTurns: [],
             runtimeRequests: [],
           },
         ],
@@ -1233,6 +1243,7 @@ describe("OrchestratorMcpService provider resolution", () => {
     messages: [],
     subagents: [],
     providerThreads: [],
+    providerTurns: [],
     runtimeRequests: [],
     turnItems: [],
   } as unknown as OrchestrationV2ThreadProjection;
@@ -3060,9 +3071,18 @@ describe("OrchestratorMcpService provider resolution", () => {
                               ordinal: 1,
                               status: "running",
                               modelSelection: runSelection,
+                              activeAttemptId: "attempt:mcp-restarted-child",
                             },
                           ]
                         : [],
+                      // Claude ran the requested Opus on Sonnet.
+                      providerTurns: [
+                        {
+                          runAttemptId: "attempt:mcp-restarted-child",
+                          ordinal: 1,
+                          model: "claude-sonnet-4-6",
+                        },
+                      ],
                     } as unknown as OrchestrationV2ThreadProjection),
               ),
           }),
@@ -3076,11 +3096,15 @@ describe("OrchestratorMcpService provider resolution", () => {
         );
         yield* Effect.gen(function* () {
           const service = yield* OrchestratorMcpService.OrchestratorMcpService;
-          for (const selection of [runSelection, threadSelection]) {
+          for (const [selection, executedModel] of [
+            [runSelection, "claude-sonnet-4-6"],
+            [threadSelection, null],
+          ] as const) {
             const result = yield* service.taskStatus(scope, taskId);
             assert.equal(result.providerInstanceId, selection.instanceId);
             assert.equal(result.model, selection.model);
             assert.deepEqual(result.options, selection.options);
+            assert.equal(result.executedModel, executedModel);
             hasRun = false;
           }
         }).pipe(Effect.provide(layerOrchestratorService.pipe(Layer.provide(dependencies))));
diff --git a/apps/server/src/mcp/OrchestratorMcpService.ts b/apps/server/src/mcp/OrchestratorMcpService.ts
index 3f91b4b3df2ab..4750e9a335f49 100644
--- a/apps/server/src/mcp/OrchestratorMcpService.ts
+++ b/apps/server/src/mcp/OrchestratorMcpService.ts
@@ -63,6 +63,7 @@ import {
   type ServerProviderModel,
   ThreadId,
   isOrchestrationV2WorkActive,
+  latestProviderTurnForAttempt,
   resolveProjectProviderInstanceEnabled,
   SECRET_REQUEST_UNSAVED_NOTE,
 } from "@t3tools/contracts";
@@ -614,7 +615,7 @@ function pageIncludesTerminalTaskResult(input: {
   readonly task: OrchestrationV2Subagent;
   readonly target: Pick<
     OrchestrationV2ThreadProjection,
-    "thread" | "runs" | "contextTransfers" | "messages" | "turnItems"
+    "thread" | "runs" | "contextTransfers" | "messages" | "turnItems" | "providerTurns"
   >;
   readonly maxChars: number;
 }): boolean {
@@ -875,13 +876,26 @@ function threadDetail(
   };
 }
 
-function threadRun(run: OrchestrationV2Run): OrchestratorMcpThreadRun {
+function runExecutedModel(
+  projection: Pick<OrchestrationV2ThreadProjection, "providerTurns">,
+  run: Pick<OrchestrationV2Run, "activeAttemptId"> | undefined,
+): string | null {
+  return run === undefined
+    ? null
+    : (latestProviderTurnForAttempt(projection.providerTurns, run.activeAttemptId)?.model ?? null);
+}
+
+function threadRun(
+  projection: Pick<OrchestrationV2ThreadProjection, "providerTurns">,
+  run: OrchestrationV2Run,
+): OrchestratorMcpThreadRun {
   return {
     runId: run.id,
     ordinal: run.ordinal,
     status: run.status,
     providerInstanceId: run.modelSelection.instanceId,
     model: run.modelSelection.model,
+    executedModel: runExecutedModel(projection, run),
     requestedAt: DateTime.formatIso(run.requestedAt),
     startedAt: run.startedAt === null ? null : DateTime.formatIso(run.startedAt),
     completedAt: run.completedAt === null ? null : DateTime.formatIso(run.completedAt),
@@ -1280,10 +1294,11 @@ const make = Effect.gen(function* () {
     const records =
       parent === undefined
         ? threadManagement
-            .getProjectThreadRecords(input, ["runs", "runtimeRequests"])
+            .getProjectThreadRecords(input, ["runs", "providerTurns", "runtimeRequests"])
             .pipe(Effect.map((target) => ({ ...target, contextTransfers: [] })))
         : threadManagement.getProjectThreadRecords(input, [
             "runs",
+            "providerTurns",
             "runtimeRequests",
             "contextTransfers",
           ]);
@@ -1509,6 +1524,7 @@ const make = Effect.gen(function* () {
           "contextTransfers",
           "subagents",
           "providerThreads",
+          "providerTurns",
           "runtimeRequests",
           "nodes",
         ],
@@ -1601,7 +1617,12 @@ const make = Effect.gen(function* () {
       : null;
     const resultContextTransferId = resultTransfer?.id ?? null;
     const latestTerminalResultContextTransferId = resultTransferForRun(terminalRun)?.id ?? null;
-    const modelSelection = childRun?.modelSelection ??
+    const executionRun =
+      task.result === null
+        ? (taskResultRun ?? childRun)
+        : (childControls.runs.find((run) => run.id === resultTransfer?.sourcePoint.runId) ??
+          childRun);
+    const modelSelection = executionRun?.modelSelection ??
       childControls.thread.modelSelection ?? {
         instanceId: task.providerInstanceId,
         model: task.model,
@@ -1620,6 +1641,7 @@ const make = Effect.gen(function* () {
       hasPendingChildRuns: hasPendingChildRuns(childProjection, childRun),
       providerInstanceId: modelSelection.instanceId,
       model: modelSelection.model,
+      executedModel: runExecutedModel(childControls, executionRun),
       options: modelSelection.options ?? [],
       summary: derivedResult,
       resultContextTransferId,
@@ -1790,6 +1812,7 @@ const make = Effect.gen(function* () {
       hasPendingChildRuns: false,
       providerInstanceId: task.providerInstanceId,
       model: task.model,
+      executedModel: null,
       summary: task.result,
       resultContextTransferId: null,
       latestTerminalRunId: null,
@@ -3130,7 +3153,7 @@ const make = Effect.gen(function* () {
         recentRuns: target.runs
           .toSorted((left, right) => right.ordinal - left.ordinal)
           .slice(0, input.runLimit ?? DEFAULT_THREAD_RUN_LIMIT)
-          .map(threadRun),
+          .map((run) => threadRun(target, run)),
         items: page.map((row) =>
           timelineItem({
             row,
diff --git a/apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.ts b/apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.ts
index 4877eed64f146..5f446dc2e1888 100644
--- a/apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.ts
+++ b/apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.ts
@@ -2970,11 +2970,15 @@ describe("ClaudeAdapterV2 background wake turns", () => {
     uuid: "00000000-0000-4000-8000-000000000101",
     session_id: WAKE_NATIVE_SESSION,
   });
-  const makeAssistantTextFrame = (input: { readonly uuid: string; readonly text: string }) =>
+  const makeAssistantTextFrame = (input: {
+    readonly uuid: string;
+    readonly text: string;
+    readonly model?: string;
+  }) =>
     claudeSdkFrame({
       type: "assistant",
       message: {
-        model: "claude-sonnet-4-6",
+        model: input.model ?? "claude-sonnet-4-6",
         id: `msg_${input.uuid}`,
         type: "message",
         role: "assistant",
@@ -6838,7 +6842,12 @@ describe("ClaudeAdapterV2 background wake turns", () => {
           }),
         );
         yield* Queue.offerAll(harness.sdkMessages, [
-          makeAssistantTextFrame({ uuid: mainUuid, text: "Delegating." }),
+          // Claude reports the model it ran, which can differ from the one requested.
+          makeAssistantTextFrame({
+            uuid: mainUuid,
+            text: "Delegating.",
+            model: "claude-opus-4-5-20251101",
+          }),
           makeAssistantErrorFrame({
             uuid: "00000000-0000-4000-8000-000000000642",
             error: undefined,
@@ -6850,6 +6859,7 @@ describe("ClaudeAdapterV2 background wake turns", () => {
 
         const providerTurn = lastProviderTurn(harness.events);
         assert.equal(providerTurn.nativeTurnRef?.nativeId, mainUuid);
+        assert.equal(providerTurn.model, "claude-opus-4-5");
         const rolledBack = yield* harness.runtime.rollbackThread({
           providerThread: harness.providerThread,
           target: {
diff --git a/apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.ts b/apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.ts
index 2824552b32ea7..6adadf385ad73 100644
--- a/apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.ts
+++ b/apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.ts
@@ -189,6 +189,7 @@ import {
   type ClaudeModelCatalog,
   getClaudeCatalogModelCapabilities,
   resolveClaudeModelSlug,
+  resolveClaudeReportedModelSlug,
   scopeClaudeModelCatalog,
 } from "../../provider/ClaudeModelCatalog.ts";
 import {
@@ -4398,6 +4399,8 @@ interface ActiveClaudeTurnContext {
   nativeWakeStarted: boolean;
   gatedFramesBeforeEcho: number;
   readonly heldRootFrames: Array<SDKMessage>;
+  // Claude can run another model than requested, such as when an organization disallows it.
+  executedModel: string | undefined;
 }
 
 interface ActiveClaudeProviderRetry {
@@ -5498,6 +5501,7 @@ export const makeClaudeAdapterV2 = Effect.fn("makeClaudeAdapterV2")(function* (
               startedAt: context.startedAt,
               completedAt: null,
               tokenUsage,
+              ...(context.executedModel === undefined ? {} : { model: context.executedModel }),
             },
           });
         });
@@ -6008,6 +6012,9 @@ export const makeClaudeAdapterV2 = Effect.fn("makeClaudeAdapterV2")(function* (
           status: input.status,
           startedAt: input.context.startedAt,
           completedAt: input.completedAt,
+          ...(input.context.executedModel === undefined
+            ? {}
+            : { model: input.context.executedModel }),
         });
 
         const buildToolCallArtifacts = (input: {
@@ -9134,6 +9141,10 @@ export const makeClaudeAdapterV2 = Effect.fn("makeClaudeAdapterV2")(function* (
               // CLI notices and API errors arrive as assistant messages from "<synthetic>".
               if (message.message.model && message.message.model !== "<synthetic>") {
                 context.rootModels.add(message.message.model);
+                context.executedModel = resolveClaudeReportedModelSlug(
+                  modelCatalog,
+                  message.message.model,
+                );
               }
               // Subagent uuids exist only in the sidechain transcript, so resume and fork cannot target them.
               context.nativeMessageCursor = message.uuid;
@@ -11262,6 +11273,7 @@ export const makeClaudeAdapterV2 = Effect.fn("makeClaudeAdapterV2")(function* (
               nativeWakeStarted: false,
               gatedFramesBeforeEcho: 0,
               heldRootFrames: [],
+              executedModel: undefined,
             };
             const attachTurn = Effect.gen(function* () {
               querySession.lastTurn = context;
diff --git a/apps/server/src/orchestration-v2/SubagentProjection.test.ts b/apps/server/src/orchestration-v2/SubagentProjection.test.ts
index ae8f7e09d2068..87ac8a2ad1758 100644
--- a/apps/server/src/orchestration-v2/SubagentProjection.test.ts
+++ b/apps/server/src/orchestration-v2/SubagentProjection.test.ts
@@ -413,6 +413,14 @@ it("orders a task's assistant messages by turn item, not by timestamps", () => {
   assert.equal(result.text, "All steps passed.");
   assert.equal(result.messageId, report.message.id);
   assert.equal(result.turnItemId, report.turnItem.id);
+  const reported = { runAttemptId: run.activeAttemptId, ordinal: 1, model: "gpt-5.4-mini" };
+  assert.equal(
+    subagentResultForRun(
+      { ...projection, messages: [report.message], providerTurns: [reported as never] },
+      run,
+    ).text,
+    "Requested model gpt-5.4; the provider reported running gpt-5.4-mini.\n\nAll steps passed.",
+  );
 });
 
 it.effect("ends an orphaned native subagent's node when only its record already ended", () =>
diff --git a/apps/server/src/peer/RemoteDelegation.test.ts b/apps/server/src/peer/RemoteDelegation.test.ts
index b487c34918cb6..317a11251aff4 100644
--- a/apps/server/src/peer/RemoteDelegation.test.ts
+++ b/apps/server/src/peer/RemoteDelegation.test.ts
@@ -100,6 +100,8 @@ interface BoxThread {
   readonly laterError: Ref.Ref<string | null>;
   /** A later run's reply, after the task's own, as a follow-up message would leave. */
   readonly laterReply: Ref.Ref<string | null>;
+  /** The model the provider there reported running the task's run. */
+  readonly executedModel: Ref.Ref<string | null>;
   /** The run each wait asked for, in order. */
   readonly waitedFor: Ref.Ref<ReadonlyArray<RunId | undefined>>;
   /** The thread's run ends with a task of its own still running, whose result wakes it again. */
@@ -145,6 +147,7 @@ const makeBoxThread = Effect.gen(function* () {
     error: yield* Ref.make<string | null>(null),
     laterError: yield* Ref.make<string | null>(null),
     laterReply: yield* Ref.make<string | null>(null),
+    executedModel: yield* Ref.make<string | null>(null),
     waitedFor: yield* Ref.make<ReadonlyArray<RunId | undefined>>([]),
     nested: yield* Ref.make(false),
     nestedWaiting: yield* Deferred.make<void>(),
@@ -226,6 +229,7 @@ const boxToolkitLayer = (thread: BoxThread) => {
       ordinal: 1,
       status,
       modelSelection,
+      activeAttemptId: "attempt:box",
       requestedAt: DateTime.makeUnsafe(0),
       startedAt: DateTime.makeUnsafe(0),
       completedAt: null,
@@ -235,9 +239,16 @@ const boxToolkitLayer = (thread: BoxThread) => {
       const shell = yield* shellOf(threadId);
       const status = yield* Ref.get(thread.status);
       const runless = yield* Ref.get(thread.runless);
+      const model = yield* Ref.get(thread.executedModel);
       return shell === null
         ? (null as never)
-        : { ...idleThreadProjection(shell), runs: runless ? [] : [runOf(status)] };
+        : {
+            ...idleThreadProjection(shell),
+            runs: runless ? [] : [runOf(status)],
+            providerTurns: (model === null
+              ? []
+              : [{ runAttemptId: "attempt:box", ordinal: 1, model }]) as never,
+          };
     });
   // One record per command id, as its receipt replays a repeat.
   const recordStop = (commandId: string, kind: "interrupt" | "thread.stop" | "delegated-tasks") =>
@@ -1264,11 +1275,14 @@ it.effect("a task's result is its own run's whole last reply, not a later run's"
       const long = "x".repeat(45_000);
       yield* Ref.set(thread.reply, long);
       yield* Ref.set(thread.laterReply, "Someone else asked a follow-up.");
+      yield* Ref.set(thread.executedModel, "gpt-5.4-mini");
       yield* Ref.set(thread.status, "completed");
       yield* Deferred.succeed(thread.finished, undefined);
       const task = yield* taskResult(a, taskId);
       expect(task?.status).toBe("completed");
-      expect(task?.result).toBe(long);
+      expect(task?.result).toBe(
+        `Requested model gpt-5.4; the provider reported running gpt-5.4-mini.\n\n${long}`,
+      );
       // Every wait was for the run the launch started.
       const waits = yield* Ref.get(thread.waitedFor);
       expect(waits.length).toBeGreaterThan(0);
diff --git a/apps/server/src/peer/RemoteDelegation.ts b/apps/server/src/peer/RemoteDelegation.ts
index 89917026ff8f7..221043f351030 100644
--- a/apps/server/src/peer/RemoteDelegation.ts
+++ b/apps/server/src/peer/RemoteDelegation.ts
@@ -5,6 +5,7 @@ import {
   OrchestratorMcpFailure,
   type OrchestratorMcpDelegateTaskInput,
   type OrchestratorMcpTaskCancelResult,
+  type OrchestratorMcpThreadRun,
   type OrchestratorMcpThreadTimelineItem,
   type OrchestrationV2RemoteTaskChild,
   type OrchestrationV2Subagent,
@@ -24,6 +25,7 @@ import * as Schedule from "effect/Schedule";
 import * as Scope from "effect/Scope";
 import * as Stream from "effect/Stream";
 import type * as Tool from "effect/ai/Tool";
+import { withExecutedModelNote } from "@t3tools/provider-core/server/subagentProjection";
 
 import type * as McpInvocationContext from "../mcp/McpInvocationContext.ts";
 import { ProjectToolkit } from "../mcp/toolkits/project/tools.ts";
@@ -435,6 +437,7 @@ const make = Effect.gen(function* () {
    * The last reply `runId` gave on the task's thread there, which is its
    * result, read whole however long it is. A failed run prefers its error
    * from the activity view over an earlier reply, without taking another run's.
+   * Led by a note when its provider reported running another model than it asked for.
    */
   const lastReply = Effect.fn("RemoteDelegation.lastReply")(function* (
     scope: ThreadScope,
@@ -445,13 +448,16 @@ const make = Effect.gen(function* () {
     let reply: OrchestratorMcpThreadTimelineItem | undefined;
     let error: OrchestratorMcpThreadTimelineItem | undefined;
     let afterPosition: number | undefined;
+    let run: OrchestratorMcpThreadRun | undefined;
     while (true) {
       const read = yield* forwarding.readThread(scope, remote.environmentId, {
         threadId: remote.threadId,
         view: failed ? "activity" : "messages",
         limit: 100,
+        runLimit: 50,
         ...(afterPosition === undefined ? {} : { afterPosition }),
       });
+      run ??= read.recentRuns.find((candidate) => candidate.runId === runId);
       reply =
         read.items.findLast(
           (item) => item.type === "assistant_message" && (runId === null || item.runId === runId),
@@ -481,7 +487,7 @@ const make = Effect.gen(function* () {
       text += part.text ?? "";
       textOffset = part.textTruncated ? (part.nextTextOffset ?? null) : null;
     }
-    return text;
+    return run === undefined ? text : withExecutedModelNote(text, run.model, run.executedModel);
   });
 
   /**
diff --git a/apps/server/src/provider/ClaudeModelCatalog.ts b/apps/server/src/provider/ClaudeModelCatalog.ts
index 57132e652ae08..4efca48f253bf 100644
--- a/apps/server/src/provider/ClaudeModelCatalog.ts
+++ b/apps/server/src/provider/ClaudeModelCatalog.ts
@@ -239,6 +239,19 @@ export function resolveClaudeModelSlug(catalog: ClaudeModelCatalog, slugOrAlias:
   return resolveClaudeCatalogModel(catalog, slugOrAlias)?.model.slug ?? slugOrAlias;
 }
 
+/** Claude reports dated IDs, such as `claude-opus-4-5-20251101`, that the catalog may not list. */
+export function resolveClaudeReportedModelSlug(
+  catalog: ClaudeModelCatalog,
+  reported: string,
+): string {
+  return (
+    (
+      resolveClaudeCatalogModel(catalog, reported) ??
+      resolveClaudeCatalogModel(catalog, reported.replace(/-\d{8}$/, ""))
+    )?.model.slug ?? reported
+  );
+}
+
 export function getClaudeCatalogModelCapabilities(
   catalog: ClaudeModelCatalog,
   slugOrAlias: string | null | undefined,
diff --git a/packages/contracts/src/orchestrationV2.ts b/packages/contracts/src/orchestrationV2.ts
index ee75d861daa7e..77a89ac3520aa 100644
--- a/packages/contracts/src/orchestrationV2.ts
+++ b/packages/contracts/src/orchestrationV2.ts
@@ -1202,6 +1202,8 @@ export const OrchestrationV2ProviderTurn = Schema.Struct({
   completedAt: Schema.NullOr(Schema.DateTimeUtc),
   tokenUsage: Schema.optional(OrchestrationV2ProviderTurnTokenUsage),
   turnTokenUsage: Schema.optional(TurnTokenUsage),
+  /** The model the provider reported running this turn's root agent, which can differ from the requested one. */
+  model: Schema.optional(TrimmedNonEmptyString),
 });
 export type OrchestrationV2ProviderTurn = typeof OrchestrationV2ProviderTurn.Type;
 
diff --git a/packages/contracts/src/orchestratorMcp.ts b/packages/contracts/src/orchestratorMcp.ts
index 5c57cd1b16fb1..d1bb9961a0c68 100644
--- a/packages/contracts/src/orchestratorMcp.ts
+++ b/packages/contracts/src/orchestratorMcp.ts
@@ -202,6 +202,12 @@ export const ORCHESTRATOR_MCP_RUNNING_TASK_NEXT =
   "Completion wakes this thread, so end your turn. Do not wait on it or poll it.";
 const OrchestratorMcpRunningTaskNext = Schema.Literal(ORCHESTRATOR_MCP_RUNNING_TASK_NEXT);
 
+// Optional: linked environments on older versions omit it.
+const OrchestratorMcpExecutedModel = Schema.NullOr(Schema.String).annotate({
+  description:
+    "Model the provider reported running, or null when unavailable. Can differ from the requested `model`.",
+});
+
 export const OrchestratorMcpDelegateTaskResult = Schema.Struct({
   taskId: NodeId,
   /** Null for a task that runs in a linked environment; see `remoteChild`. */
@@ -222,6 +228,7 @@ export const OrchestratorMcpDelegateTaskResult = Schema.Struct({
   latestTerminalResultContextTransferId: Schema.NullOr(ContextTransferId),
   providerInstanceId: ProviderInstanceId,
   model: Schema.NullOr(Schema.String),
+  executedModel: Schema.optional(OrchestratorMcpExecutedModel),
   options: Schema.optional(Schema.Array(ProviderOptionSelection)),
   summary: Schema.NullOr(Schema.String),
   resultContextTransferId: Schema.NullOr(ContextTransferId),
@@ -463,6 +470,7 @@ export const OrchestratorMcpThreadRun = Schema.Struct({
   status: OrchestrationV2RunStatus,
   providerInstanceId: ProviderInstanceId,
   model: Schema.String,
+  executedModel: Schema.optional(OrchestratorMcpExecutedModel),
   requestedAt: IsoDateTime,
   startedAt: Schema.NullOr(IsoDateTime),
   completedAt: Schema.NullOr(IsoDateTime),
diff --git a/packages/provider-core/src/server/subagentProjection.ts b/packages/provider-core/src/server/subagentProjection.ts
index dbd929192e04d..62145567f6651 100644
--- a/packages/provider-core/src/server/subagentProjection.ts
+++ b/packages/provider-core/src/server/subagentProjection.ts
@@ -22,7 +22,7 @@ import { backgroundWorkHoldsCompletion } from "@t3tools/shared/orchestrationV2Pe
 import { runRanAfter } from "@t3tools/shared/orchestrationV2ThreadError";
 import * as DateTime from "effect/DateTime";
 import * as Effect from "effect/Effect";
-import { isOrchestrationV2WorkActive } from "@t3tools/contracts";
+import { isOrchestrationV2WorkActive, latestProviderTurnForAttempt } from "@t3tools/contracts";
 
 function trimmed(value: string | null | undefined): string | undefined {
   const result = value?.trim();
@@ -176,8 +176,19 @@ export function makeSubagentConversationArtifacts(input: {
   return { message, turnItem };
 }
 
+/** A delegated result, led by a note when the provider reported running another model than requested. */
+export function withExecutedModelNote(
+  result: string,
+  requestedModel: string,
+  executedModel: string | null | undefined,
+): string {
+  return executedModel == null || executedModel === requestedModel
+    ? result
+    : `Requested model ${requestedModel}; the provider reported running ${executedModel}.\n\n${result}`;
+}
+
 export function subagentResultForRun(
-  projection: Pick<OrchestrationV2ThreadProjection, "messages" | "turnItems">,
+  projection: Pick<OrchestrationV2ThreadProjection, "messages" | "turnItems" | "providerTurns">,
   run: OrchestrationV2Run,
 ): {
   readonly text: string;
@@ -223,8 +234,15 @@ export function subagentResultForRun(
     (run.status === "completed"
       ? "Child task completed without an assistant result."
       : `Child task ended with status ${run.status}.`);
+  const executedModel = latestProviderTurnForAttempt(
+    projection.providerTurns,
+    run.activeAttemptId,
+  )?.model;
   return {
-    text,
+    text:
+      executedModel === undefined
+        ? text
+        : withExecutedModelNote(text, run.modelSelection.model, executedModel),
     messageId: failure === undefined ? (message?.id ?? turnItem?.messageId ?? null) : null,
     turnItemId: failure?.id ?? turnItem?.id ?? null,
   };