From cad7e7f9e1eef2ff20a3533d3be5026de5a5dfa4 Mon Sep 17 00:00:00 2001
From: Peter Steinberger <steipete@gmail.com>
Date: Wed, 12 Aug 2026 19:46:56 -0700
Subject: [PATCH] fix: session model selection replaced by response aliases and
 fallback wins (#122910)

* fix(agents): keep session model selection canonical

Preserve configured session model identity across response aliases and turn-local fallbacks.

* fix(gateway): align session model projection consumers

* test(gateway): mark session event model selection
---
 .../agent-command.live-model-switch.test.ts   |  3 +
 src/agents/command/post-run.ts                |  2 +
 src/agents/command/session-store.ts           |  9 ++-
 .../run/terminal-preparation.test.ts          |  6 +-
 .../run/terminal-preparation.ts               |  5 +-
 .../reply/agent-runner-result-accounting.ts   |  3 +-
 .../agent-runner.runreplyagent.e2e.test.ts    | 20 +++---
 .../server.chat.gateway-server-chat-b.test.ts |  4 +-
 .../server.sessions.list-changed.test.ts      | 10 ++-
 .../server.sessions.thinking-e2e.test.ts      | 63 +++++++++++++++++++
 src/gateway/session-message-events.test.ts    |  2 +
 src/gateway/session-utils-projection.ts       |  7 +--
 src/gateway/session-utils-row.ts              | 33 +++-------
 src/gateway/session-utils-search.ts           | 15 +----
 src/gateway/session-utils.search.test.ts      | 25 +++++---
 src/gateway/session-utils.test.ts             | 16 +++--
 16 files changed, 143 insertions(+), 80 deletions(-)

diff --git a/src/agents/agent-command.live-model-switch.test.ts b/src/agents/agent-command.live-model-switch.test.ts
index cd059949228c..a022be7f2e66 100644
--- a/src/agents/agent-command.live-model-switch.test.ts
+++ b/src/agents/agent-command.live-model-switch.test.ts
@@ -2445,6 +2445,9 @@ describe("agentCommand – LiveSessionModelSwitchError retry", () => {
         entry: expect.objectContaining({ thinkingLevel: "max" }),
       }),
     );
+    expectRecordFields(mockCallArg(state.updateSessionStoreAfterAgentRunMock), {
+      preserveRuntimeModel: true,
+    });
   });
 
   it("recomputes a model-derived thinking default for each fallback candidate", async () => {
diff --git a/src/agents/command/post-run.ts b/src/agents/command/post-run.ts
index 130d23575e21..2b1e8ff0e6b6 100644
--- a/src/agents/command/post-run.ts
+++ b/src/agents/command/post-run.ts
@@ -165,6 +165,8 @@ export async function finalizeEmbeddedAgentCommand(params: {
         touchActivity: !isHeartbeatLifecycleRun && !params.opts.internalEvents?.length,
         preserveRuntimeModel:
           fallbackExhausted ||
+          fallbackProvider !== provider ||
+          fallbackModel !== model ||
           isHeartbeatLifecycleRun ||
           params.preserveUserFacingSessionModelState,
         preserveUserFacingSessionModelState: params.preserveUserFacingSessionModelState,
diff --git a/src/agents/command/session-store.ts b/src/agents/command/session-store.ts
index 6ae227e6b2a1..121eb2fe1b66 100644
--- a/src/agents/command/session-store.ts
+++ b/src/agents/command/session-store.ts
@@ -66,8 +66,8 @@ export async function updateSessionStoreAfterAgentRun(params: {
   /**
    * When true, preserve the pre-existing runtime model fields (model,
    * modelProvider, contextTokens) on the session entry instead of overwriting
-   * them with the model used by this run. Used for heartbeat turns so the
-   * heartbeat model does not "bleed" into the main session's perceived state.
+   * them with the model used by this run. Used for turn-local fallback and
+   * heartbeat runs so their model does not bleed into the session selection.
    */
   preserveRuntimeModel?: boolean;
   preserveUserFacingSessionModelState?: boolean;
@@ -146,9 +146,8 @@ export async function updateSessionStoreAfterAgentRun(params: {
     );
   }
   if (preserveRuntimeModel) {
-    // Keep the pre-existing runtime model and context window so a background
-    // heartbeat turn using a different model does not bleed into the main
-    // session's perceived state.
+    // Keep the pre-existing runtime model and context window so a turn-local
+    // model does not bleed into the session's perceived selection.
     if (entry.model) {
       // Prior runtime model exists: preserve its contextTokens. When missing,
       // leave contextTokens unset rather than falling back to the heartbeat
diff --git a/src/agents/embedded-agent-runner/run/terminal-preparation.test.ts b/src/agents/embedded-agent-runner/run/terminal-preparation.test.ts
index 89888064dd20..928467d87c38 100644
--- a/src/agents/embedded-agent-runner/run/terminal-preparation.test.ts
+++ b/src/agents/embedded-agent-runner/run/terminal-preparation.test.ts
@@ -251,7 +251,7 @@ describe("prepareEmbeddedRunTerminal run stats", () => {
     expect(prepared.agentMeta).not.toHaveProperty("costUsd");
   });
 
-  it("builds exact terminal model and successful-tool evidence", async () => {
+  it("keeps response identity in the terminal receipt without replacing the run model", async () => {
     const prepared = await prepareStats({
       responseModel: "cost-model-rerouted",
       attempt: {
@@ -276,7 +276,7 @@ describe("prepareEmbeddedRunTerminal run stats", () => {
       requested: { provider: "cost-test-provider", model: "cost-model" },
       effective: {
         provider: "cost-test-provider",
-        model: "cost-model-rerouted",
+        model: "cost-model",
         responseModel: "cost-model-rerouted",
       },
       successfulToolNames: ["exec", "read", "Zeta", "alpha", "zeta"],
@@ -285,5 +285,7 @@ describe("prepareEmbeddedRunTerminal run stats", () => {
     expect(
       (prepared.agentMeta as { terminalReceipt?: Record<string, unknown> }).terminalReceipt,
     ).not.toHaveProperty("terminalDisposition");
+    expect(prepared.agentMeta.model).toBe("cost-model");
+    expect(prepared.reportedModelRef.model).toBe("cost-model");
   });
 });
diff --git a/src/agents/embedded-agent-runner/run/terminal-preparation.ts b/src/agents/embedded-agent-runner/run/terminal-preparation.ts
index a44bc5eb267b..e6e3f10f7137 100644
--- a/src/agents/embedded-agent-runner/run/terminal-preparation.ts
+++ b/src/agents/embedded-agent-runner/run/terminal-preparation.ts
@@ -75,13 +75,12 @@ export function prepareEmbeddedRunTerminal(input: {
     latestUsage: terminalAssistant?.usage as UsageLike | undefined,
     lastRunPromptUsage: input.lastRunPromptUsage,
   });
-  const resolvedModelRef = resolveReportedModelRef({
+  const reportedModelRef = resolveReportedModelRef({
     provider: input.provider,
     model: input.model,
     assistant: terminalAssistant,
   });
-  const responseModel = terminalAssistant?.responseModel?.trim() || resolvedModelRef.model;
-  const reportedModelRef = { ...resolvedModelRef, model: responseModel };
+  const responseModel = terminalAssistant?.responseModel?.trim() || reportedModelRef.model;
   const finalAssistantStopReason = (terminalAssistant?.stopReason ?? "").trim().toLowerCase();
   const terminalAssistantCanOwnFinalText =
     finalAssistantStopReason !== "error" && finalAssistantStopReason !== "aborted";
diff --git a/src/auto-reply/reply/agent-runner-result-accounting.ts b/src/auto-reply/reply/agent-runner-result-accounting.ts
index 0163a44e0be6..9aa2119b91af 100644
--- a/src/auto-reply/reply/agent-runner-result-accounting.ts
+++ b/src/auto-reply/reply/agent-runner-result-accounting.ts
@@ -253,7 +253,8 @@ export async function accountAgentTurn(context: AgentTurnAccountingContext) {
     compactionTokensAfter: runResult.meta?.agentMeta?.compactionTokensAfter,
     promptTokens,
     isHeartbeat,
-    preserveRuntimeModel: fallbackExhausted,
+    preserveRuntimeModel:
+      fallbackExhausted || fallbackTransition.nextState.selectedModel !== undefined,
     preserveUserFacingSessionModelState: preserveUserFacingSessionState,
     modelUsed,
     providerUsed,
diff --git a/src/auto-reply/reply/agent-runner.runreplyagent.e2e.test.ts b/src/auto-reply/reply/agent-runner.runreplyagent.e2e.test.ts
index 7ec9d3dd2d67..60a4141a13c6 100644
--- a/src/auto-reply/reply/agent-runner.runreplyagent.e2e.test.ts
+++ b/src/auto-reply/reply/agent-runner.runreplyagent.e2e.test.ts
@@ -3439,8 +3439,10 @@ describe("runReplyAgent typing (heartbeat)", () => {
     for (const testCase of cases) {
       const sessionEntry = makeSessionEntry({
         providerOverride: "openai",
-        modelOverride: "gpt-5.6-luna",
+        modelOverride: "gpt-5.6-sol",
         modelOverrideSource: "user",
+        modelProvider: "openai",
+        model: "gpt-5.6-sol",
       });
       await replaceSessionEntry({ storePath, sessionKey: "main" }, sessionEntry);
       const sessionStore = { main: sessionEntry };
@@ -3451,10 +3453,10 @@ describe("runReplyAgent typing (heartbeat)", () => {
       vi.spyOn(modelFallbackModule, "runWithModelFallback").mockImplementationOnce(async (args) => {
         const { run, onFallbackStep } = args;
         expect(args.provider, testCase.name).toBe("openai");
-        expect(args.model, testCase.name).toBe("gpt-5.6-luna");
+        expect(args.model, testCase.name).toBe("gpt-5.6-sol");
         await onFallbackStep?.({
           fallbackStepType: "fallback_step",
-          fallbackStepFromModel: "openai/gpt-5.6-luna",
+          fallbackStepFromModel: "openai/gpt-5.6-sol",
           fallbackStepToModel: "deepinfra/moonshotai/Kimi-K2.5",
           fallbackStepFromFailureReason: "rate_limit",
           fallbackStepFinalOutcome: "succeeded",
@@ -3467,7 +3469,7 @@ describe("runReplyAgent typing (heartbeat)", () => {
           attempts: [
             {
               provider: "openai",
-              model: "gpt-5.6-luna",
+              model: "gpt-5.6-sol",
               error: "Provider openai is in cooldown (all profiles unavailable)",
               reason: "rate_limit",
             },
@@ -3481,7 +3483,7 @@ describe("runReplyAgent typing (heartbeat)", () => {
         sessionStore,
         sessionKey: "main",
         storePath,
-        runOverrides: { provider: "openai", model: "gpt-5.6-luna" },
+        runOverrides: { provider: "openai", model: "gpt-5.6-sol" },
       });
       const phases: string[] = [];
       const off = onAgentEvent((evt) => {
@@ -3499,11 +3501,11 @@ describe("runReplyAgent typing (heartbeat)", () => {
       expect(payload.text, testCase.name).toContain("Model Fallback:");
       expect(payload.text, testCase.name).toContain("deepinfra/moonshotai/Kimi-K2.5");
       expect(stored.providerOverride, testCase.name).toBe("openai");
-      expect(stored.modelOverride, testCase.name).toBe("gpt-5.6-luna");
+      expect(stored.modelOverride, testCase.name).toBe("gpt-5.6-sol");
       expect(stored.modelOverrideSource, testCase.name).toBe("user");
-      expect(stored.modelProvider, testCase.name).toBe("deepinfra");
-      expect(stored.model, testCase.name).toBe("moonshotai/Kimi-K2.5");
-      expect(stored.fallbackNotice?.selectedModel, testCase.name).toBe("openai/gpt-5.6-luna");
+      expect(stored.modelProvider, testCase.name).toBe("openai");
+      expect(stored.model, testCase.name).toBe("gpt-5.6-sol");
+      expect(stored.fallbackNotice?.selectedModel, testCase.name).toBe("openai/gpt-5.6-sol");
       expect(stored.fallbackNotice?.activeModel, testCase.name).toBe(
         "deepinfra/moonshotai/Kimi-K2.5",
       );
diff --git a/src/gateway/server.chat.gateway-server-chat-b.test.ts b/src/gateway/server.chat.gateway-server-chat-b.test.ts
index 427ca56b881b..64a750dc05a4 100644
--- a/src/gateway/server.chat.gateway-server-chat-b.test.ts
+++ b/src/gateway/server.chat.gateway-server-chat-b.test.ts
@@ -903,13 +903,15 @@ describe("gateway server chat", () => {
     }
   });
 
-  test("chat.history exposes persisted and synthetic session metadata for startup hydration", async () => {
+  test("chat.history exposes selected and synthetic session metadata for startup hydration", async () => {
     await withGatewayChatHarness(async ({ ws, createSessionDir }) => {
       await connectOk(ws);
       await createSessionDir();
       const updatedAt = Date.now();
       await writeStoredMainSession({
         updatedAt,
+        providerOverride: "openai",
+        modelOverride: "gpt-5",
         modelProvider: "openai",
         model: "gpt-5",
         contextTokens: 128_000,
diff --git a/src/gateway/server.sessions.list-changed.test.ts b/src/gateway/server.sessions.list-changed.test.ts
index 5a2af4a1b625..321ed642a193 100644
--- a/src/gateway/server.sessions.list-changed.test.ts
+++ b/src/gateway/server.sessions.list-changed.test.ts
@@ -261,7 +261,7 @@ async function expectListedSessionActiveRun(
   expect(session.activeRunIds).toEqual(expected ? ["run-1"] : undefined);
 }
 
-test("sessions.list keeps bulk rows lightweight and uses persisted model fields", async () => {
+test("sessions.list keeps bulk rows lightweight and uses selected model fields", async () => {
   const { storePath } = await createSessionStoreDir();
   testState.agentConfig = {
     models: {
@@ -273,6 +273,8 @@ test("sessions.list keeps bulk rows lightweight and uses persisted model fields"
       main: sessionStoreEntry("sess-parent"),
       "dashboard:child": sessionStoreEntry("sess-child", {
         updatedAt: Date.now() - 1_000,
+        providerOverride: "anthropic",
+        modelOverride: "test-model-without-catalog-context",
         modelProvider: "anthropic",
         model: "test-model-without-catalog-context",
         modelSelectionLocked: true,
@@ -351,13 +353,15 @@ test.each([
   ["my-ngc", "deepseek-ai/deepseek-v4-pro"],
   ["my-ngc:nvidia", "nvidia/nemotron-3-ultra-550b-a55b"],
 ])(
-  "sessions.list preserves custom provider %s and nested models over WebSocket",
+  "sessions.list preserves selected custom provider %s and nested models over WebSocket",
   async (provider, model) => {
     const { storePath } = await createSessionStoreDir();
     await writeSessionStore({
       entries: {
         main: sessionStoreEntry("sess-parent"),
         "dashboard:child": sessionStoreEntry("sess-custom-provider", {
+          providerOverride: provider,
+          modelOverride: model,
           modelProvider: provider,
           model,
           parentSessionKey: "agent:main:main",
@@ -857,6 +861,8 @@ test("sessions.changed mutation events include live usage metadata", async () =>
   await writeSessionStore({
     entries: {
       main: sessionStoreEntry("sess-main", {
+        providerOverride: "openai",
+        modelOverride: "gpt-5.3-codex-spark",
         modelProvider: "openai",
         model: "gpt-5.3-codex-spark",
         contextTokens: 123_456,
diff --git a/src/gateway/server.sessions.thinking-e2e.test.ts b/src/gateway/server.sessions.thinking-e2e.test.ts
index e5b9d2ff91b0..b1b26ff7b168 100644
--- a/src/gateway/server.sessions.thinking-e2e.test.ts
+++ b/src/gateway/server.sessions.thinking-e2e.test.ts
@@ -12,6 +12,7 @@ import { expect, test, vi } from "vitest";
 import { formatThinkingLevels } from "../auto-reply/thinking.js";
 import { testState, writeSessionStore } from "./test-helpers.js";
 import {
+  directSessionReq,
   setupGatewaySessionsHandlerTestHarness,
   getGatewayConfigModule,
   getSessionsHandlers,
@@ -77,6 +78,7 @@ type ThinkingSession = {
   key: string;
   modelProvider?: string;
   model?: string;
+  agentRuntime?: { id?: string };
   thinkingLevels?: Array<{ label: string }>;
   thinkingOptions?: string[];
 };
@@ -91,6 +93,8 @@ async function listMainSessionWithThinking(params: {
   primaryModel: string;
   sessionModelProvider: string;
   sessionModel: string;
+  agentRuntime?: "codex" | "openclaw";
+  selectedByOverride?: boolean;
   readPreparedGatewayModelCatalog?: () => Promise<
     Array<{
       provider: string;
@@ -104,12 +108,25 @@ async function listMainSessionWithThinking(params: {
   await createSessionStoreDir();
   testState.agentConfig = {
     model: { primary: params.primaryModel },
+    ...(params.agentRuntime
+      ? {
+          models: {
+            [params.primaryModel]: { agentRuntime: { id: params.agentRuntime } },
+          },
+        }
+      : {}),
   };
   await writeSessionStore({
     entries: {
       main: sessionStoreEntry("sess-main", {
         modelProvider: params.sessionModelProvider,
         model: params.sessionModel,
+        ...(params.selectedByOverride === false
+          ? {}
+          : {
+              providerOverride: params.sessionModelProvider,
+              modelOverride: params.sessionModel,
+            }),
       }),
     },
   });
@@ -211,3 +228,49 @@ test("e2e #76482: session matching default model inherits default thinking level
   expect(resolved).toContain("off");
   expect(resolved).toContain("high");
 });
+
+test("session rows keep the selected Codex Sol model when runtime metadata contains a response family", async () => {
+  const loadSolCatalog = async () => [
+    {
+      provider: "openai",
+      id: "gpt-5.6-sol",
+      name: "GPT-5.6-Sol",
+      reasoning: true,
+      compat: {
+        supportedReasoningEfforts: ["low", "medium", "high", "xhigh", "max", "ultra"],
+      },
+    },
+  ];
+  const { session } = await listMainSessionWithThinking({
+    reqId: "req-e2e-codex-sol-family",
+    primaryModel: "openai/gpt-5.6-sol",
+    sessionModelProvider: "openai",
+    sessionModel: "gpt-5.6",
+    agentRuntime: "codex",
+    selectedByOverride: false,
+    readPreparedGatewayModelCatalog: loadSolCatalog,
+  });
+
+  expect(session).toMatchObject({
+    modelProvider: "openai",
+    model: "gpt-5.6-sol",
+  });
+  expect(session?.agentRuntime?.id).toBe("codex");
+  expect(session?.thinkingOptions).toContain("max");
+
+  const patchResponse = await directSessionReq(
+    "sessions.patch",
+    { key: "main", thinkingLevel: "max" },
+    { context: { loadGatewayModelCatalog: loadSolCatalog } },
+  );
+  expect(patchResponse.ok, patchResponse.error?.message).toBe(true);
+  expect(patchResponse.error).toBeUndefined();
+  expect(patchResponse.payload).toMatchObject({
+    ok: true,
+    resolved: {
+      modelProvider: "openai",
+      model: "gpt-5.6-sol",
+      thinkingLevel: "max",
+    },
+  });
+});
diff --git a/src/gateway/session-message-events.test.ts b/src/gateway/session-message-events.test.ts
index 094ea73ef0a2..6da9a686d0aa 100644
--- a/src/gateway/session-message-events.test.ts
+++ b/src/gateway/session-message-events.test.ts
@@ -1753,6 +1753,8 @@ describe("session.message websocket events", () => {
         main: {
           sessionId: "sess-main",
           updatedAt: Date.now(),
+          providerOverride: "openai",
+          modelOverride: "gpt-5.4",
           modelProvider: "openai",
           model: "gpt-5.4",
           contextTokens: 123_456,
diff --git a/src/gateway/session-utils-projection.ts b/src/gateway/session-utils-projection.ts
index c9aea3e65f90..a85f66731a40 100644
--- a/src/gateway/session-utils-projection.ts
+++ b/src/gateway/session-utils-projection.ts
@@ -87,14 +87,11 @@ export function resolveSessionSelectedModelRef(params: {
   agentId: string;
   rowContext?: SessionListRowContext;
   allowPluginNormalization?: boolean;
-}): ReturnType<typeof resolveSessionModelRef> | null {
+}): ReturnType<typeof resolveSessionModelRef> {
   const override = normalizeStoredOverrideModel({
     providerOverride: params.entry?.providerOverride,
     modelOverride: params.entry?.modelOverride,
   });
-  if (!override.modelOverride) {
-    return null;
-  }
   if (!params.rowContext) {
     return resolveSessionModelRef(params.cfg, params.entry, params.agentId, {
       allowPluginNormalization: params.allowPluginNormalization,
@@ -103,7 +100,7 @@ export function resolveSessionSelectedModelRef(params: {
   const key = [
     normalizeAgentId(params.agentId),
     override.providerOverride ?? "",
-    override.modelOverride,
+    override.modelOverride ?? "",
   ].join("\0");
   const cached = params.rowContext.selectedModelByOverrideRef.get(key);
   if (cached) {
diff --git a/src/gateway/session-utils-row.ts b/src/gateway/session-utils-row.ts
index 98829a8603ef..5e4a1fd7b31e 100644
--- a/src/gateway/session-utils-row.ts
+++ b/src/gateway/session-utils-row.ts
@@ -234,8 +234,6 @@ export function buildGatewaySessionRow(params: {
     subagentRun?.model,
     { allowPluginNormalization: !lightweight },
   );
-  const runtimeModelPresent =
-    Boolean(entry?.model?.trim()) || Boolean(entry?.modelProvider?.trim());
   const freshSessionTotalTokens = asNonNegativeFiniteNumber(resolveFreshSessionTotalTokens(entry));
   const needsTranscriptTotalTokens = freshSessionTotalTokens === undefined;
   const needsTranscriptContextTokens = resolvePositiveNumber(entry?.contextTokens) === undefined;
@@ -263,23 +261,6 @@ export function buildGatewaySessionRow(params: {
           agentId: sessionAgentId,
         })
       : null;
-  const preferLiveSubagentModelIdentity =
-    Boolean(subagentRun?.model?.trim()) && subagentStatus === "running";
-  const shouldUseTranscriptModelIdentity =
-    runtimeModelPresent &&
-    !preferLiveSubagentModelIdentity &&
-    (needsTranscriptTotalTokens || needsTranscriptContextTokens);
-  const resolvedModelIdentity = {
-    provider: resolvedModel.provider,
-    model: resolvedModel.model ?? DEFAULT_MODEL,
-  };
-  const modelIdentity = shouldUseTranscriptModelIdentity
-    ? {
-        provider: transcriptUsage?.modelProvider ?? resolvedModelIdentity.provider,
-        model: transcriptUsage?.model ?? resolvedModelIdentity.model,
-      }
-    : resolvedModelIdentity;
-  const { provider: modelProvider, model } = modelIdentity;
   const totalTokens =
     freshSessionTotalTokens ?? asNonNegativeFiniteNumber(transcriptUsage?.totalTokens);
   const totalTokensFresh =
@@ -314,15 +295,15 @@ export function buildGatewaySessionRow(params: {
   const latestCompactionCheckpoint = buildCompactionCheckpointPreview(
     resolveLatestCompactionCheckpoint(compactionCheckpoints),
   );
-  const selectedOrRuntimeModelProvider = selectedModel?.provider ?? modelProvider;
-  const selectedOrRuntimeModel = selectedModel?.model ?? model;
+  const selectedModelProvider = selectedModel.provider;
+  const selectedModelId = selectedModel.model;
   const rowModelIdentity = lightweight
-    ? { provider: selectedOrRuntimeModelProvider, model: selectedOrRuntimeModel }
+    ? { provider: selectedModelProvider, model: selectedModelId }
     : resolveSessionDisplayModelIdentityRefCached({
         cfg,
         agentId: sessionAgentId,
-        provider: selectedOrRuntimeModelProvider,
-        model: selectedOrRuntimeModel,
+        provider: selectedModelProvider,
+        model: selectedModelId,
         rowContext: params.rowContext,
       });
   const rowModelProvider = rowModelIdentity.provider;
@@ -394,8 +375,8 @@ export function buildGatewaySessionRow(params: {
   });
   const fastModeState = resolveFastModeState({
     cfg,
-    provider: selectedOrRuntimeModelProvider ?? DEFAULT_PROVIDER,
-    model: selectedOrRuntimeModel ?? DEFAULT_MODEL,
+    provider: selectedModelProvider,
+    model: selectedModelId,
     agentId: sessionAgentId,
     sessionEntry:
       entry?.fastMode !== undefined
diff --git a/src/gateway/session-utils-search.ts b/src/gateway/session-utils-search.ts
index e74f5a436ef7..edd7050ffd8a 100644
--- a/src/gateway/session-utils-search.ts
+++ b/src/gateway/session-utils-search.ts
@@ -2,7 +2,6 @@ import {
   normalizeLowercaseStringOrEmpty,
   normalizeOptionalString,
 } from "@openclaw/normalization-core/string-coerce";
-import { DEFAULT_MODEL } from "../agents/defaults.js";
 import type { ModelCatalogEntry } from "../agents/model-catalog.js";
 import { resolveSessionModelIdentityRef } from "../agents/session-model-ref.js";
 import { getSessionDisplaySubagentRunByChildSessionKey } from "../agents/subagents/registry/subagent-registry-read.js";
@@ -128,17 +127,11 @@ export function resolveSessionListSearchModelFields(params: {
     subagentRun?.model,
     { allowPluginNormalization: false },
   );
-  const modelIdentity = {
-    provider: resolvedModel.provider,
-    model: resolvedModel.model ?? DEFAULT_MODEL,
-  };
-  const selectedOrRuntimeModelProvider = selectedModel?.provider ?? modelIdentity.provider;
-  const selectedOrRuntimeModel = selectedModel?.model ?? modelIdentity.model;
   const displayModelIdentity = resolveSessionDisplayModelIdentityRefCached({
     cfg: params.cfg,
     agentId,
-    provider: selectedOrRuntimeModelProvider,
-    model: selectedOrRuntimeModel,
+    provider: selectedModel.provider,
+    model: selectedModel.model,
     rowContext: params.rowContext,
   });
   const fields: Array<string | undefined> = [];
@@ -147,9 +140,7 @@ export function resolveSessionListSearchModelFields(params: {
     model: params.entry?.model,
   });
   addSessionListSearchModelFields(fields, resolvedModel);
-  if (selectedModel) {
-    addSessionListSearchModelFields(fields, selectedModel);
-  }
+  addSessionListSearchModelFields(fields, selectedModel);
   addSessionListSearchModelFields(fields, displayModelIdentity);
   return fields;
 }
diff --git a/src/gateway/session-utils.search.test.ts b/src/gateway/session-utils.search.test.ts
index 8be88ac690a0..a8478304f43d 100644
--- a/src/gateway/session-utils.search.test.ts
+++ b/src/gateway/session-utils.search.test.ts
@@ -96,7 +96,7 @@ describe("filterAndSortSessionEntries search", () => {
     }
   });
 
-  test("filters by displayed provider and model identity", () => {
+  test("filters by selected and stored provider and model identity", () => {
     const now = Date.now();
     const cfg = createModelDefaultsConfig("anthropic/claude-sonnet-4-6");
     const store: Record<string, SessionEntry> = {
@@ -121,12 +121,21 @@ describe("filterAndSortSessionEntries search", () => {
       } as SessionEntry,
     };
     const cases = [
-      { search: "anthropic", expectedKey: "agent:main:inherited-default" },
-      { search: "claude-sonnet", expectedKey: "agent:main:inherited-default" },
-      { search: "anthropic/claude-sonnet", expectedKey: "agent:main:inherited-default" },
-      { search: "openai/gpt-5.5", expectedKey: "agent:main:override" },
-      { search: "gemini-3.1", expectedKey: "agent:main:runtime" },
-      { search: "google/gemini", expectedKey: "agent:main:runtime" },
+      {
+        search: "anthropic",
+        expectedKeys: ["agent:main:inherited-default", "agent:main:runtime"],
+      },
+      {
+        search: "claude-sonnet",
+        expectedKeys: ["agent:main:inherited-default", "agent:main:runtime"],
+      },
+      {
+        search: "anthropic/claude-sonnet",
+        expectedKeys: ["agent:main:inherited-default", "agent:main:runtime"],
+      },
+      { search: "openai/gpt-5.5", expectedKeys: ["agent:main:override"] },
+      { search: "gemini-3.1", expectedKeys: ["agent:main:runtime"] },
+      { search: "google/gemini", expectedKeys: ["agent:main:runtime"] },
     ] as const;
 
     for (const testCase of cases) {
@@ -137,7 +146,7 @@ describe("filterAndSortSessionEntries search", () => {
           opts: { search: testCase.search },
           now,
         }),
-      ).toEqual([testCase.expectedKey]);
+      ).toEqual(testCase.expectedKeys);
     }
   });
 
diff --git a/src/gateway/session-utils.test.ts b/src/gateway/session-utils.test.ts
index a8093f98c334..2357850fa858 100644
--- a/src/gateway/session-utils.test.ts
+++ b/src/gateway/session-utils.test.ts
@@ -916,12 +916,16 @@ describe("gateway session utils", () => {
       store: {
         upper: {
           sessionId: "upper",
+          providerOverride: "custom",
+          modelOverride: "CaseModel",
           modelProvider: "custom",
           model: "CaseModel",
           updatedAt: 2,
         } satisfies SessionEntry,
         lower: {
           sessionId: "lower",
+          providerOverride: "custom",
+          modelOverride: "casemodel",
           modelProvider: "custom",
           model: "casemodel",
           updatedAt: 1,
@@ -3359,7 +3363,7 @@ describe("listSessionsFromStore selected model display", () => {
     });
   });
 
-  test("infers canonical provider for bare CLI models before default-provider fallback", () => {
+  test("ignores bare CLI runtime metadata when the selected default differs", () => {
     const cfg = createModelDefaultsConfig({
       primary: "openai/gpt-5.4",
       models: {
@@ -3382,8 +3386,8 @@ describe("listSessionsFromStore selected model display", () => {
       opts: {},
     });
 
-    expect(result.sessions[0]?.modelProvider).toBe("anthropic");
-    expect(result.sessions[0]?.model).toBe("claude-opus-4-7");
+    expect(result.sessions[0]?.modelProvider).toBe("openai");
+    expect(result.sessions[0]?.model).toBe("gpt-5.4");
   });
 
   test("uses qualified selected defaults for rows without runtime model metadata", () => {
@@ -3435,7 +3439,7 @@ describe("listSessionsFromStore selected model display", () => {
     ]);
   });
 
-  test("uses persisted runtime model metadata before selected defaults", () => {
+  test("uses selected defaults before persisted runtime model metadata", () => {
     const cfg = {
       agents: {
         defaults: { model: { primary: "openai/gpt-5.4" } },
@@ -3457,8 +3461,8 @@ describe("listSessionsFromStore selected model display", () => {
       opts: {},
     });
 
-    expect(result.sessions[0]?.modelProvider).toBe("openai");
-    expect(result.sessions[0]?.model).toBe("gpt-5.5");
+    expect(result.sessions[0]?.modelProvider).toBe("anthropic");
+    expect(result.sessions[0]?.model).toBe("claude-sonnet-4-6");
   });
 
   test("uses complete model overrides without default-model fallback", () => {