fix: clarify token and video usage in model call logs and prevent retry messages from inheriting token usage

backend/internal/application/billing/service.go

@@ -151,6 +151,8 @@ type UsagePricingInput struct {
 	CallCount           int64
 	DurationBillable    bool
 	DurationSeconds     int64
+	MediaType           string
+	InputImageCount     int64
 	LatencyMS           int64
 	ServerSideToolUsage map[string]int64
 	ServiceItems        []ServiceUsageInput
@@ -1768,6 +1770,14 @@ func (s *Service) BuildUsageLedger(ctx context.Context, input UsagePricingInput)
 		"base_service_billed_nanousd":              serviceBilledNanousd,
 		"service_items":                            usageServiceItemSnapshots(serviceItems),
 	}
+	if strings.EqualFold(strings.TrimSpace(input.MediaType), "video") {
+		inputImageCount := input.InputImageCount
+		if inputImageCount < 0 {
+			inputImageCount = 0
+		}
+		snapshot["media_type"] = "video"
+		snapshot["input_image_count"] = inputImageCount
+	}
 	if usageSource := strings.TrimSpace(input.UsageSource); usageSource != "" {
 		snapshot["usage_source"] = usageSource
 	}

backend/internal/application/billing/service_model_identity_test.go

@@ -101,13 +101,22 @@ func TestBuildUsageLedgerBillsDurationOnlyWhenExplicitlyBillable(t *testing.T) {
 		PlatformModelName: "video-model",
 		DurationBillable:  true,
 		DurationSeconds:   6,
+		MediaType:         "video",
+		InputImageCount:   1,
 	})
 	if err != nil {
 		t.Fatalf("build video duration ledger: %v", err)
 	}
 	if video.DurationSeconds != 6 || video.BilledNanousd != 18 {
 		t.Fatalf("unexpected video duration billing: %#v", video)
 	}
+	var snapshot map[string]interface{}
+	if err := json.Unmarshal([]byte(video.PricingSnapshotJSON), &snapshot); err != nil {
+		t.Fatalf("unmarshal video pricing snapshot: %v", err)
+	}
+	if snapshot["media_type"] != "video" || snapshot["input_image_count"] != float64(1) {
+		t.Fatalf("unexpected video media snapshot: %#v", snapshot)
+	}
 }
 
 func TestAuthorizeUsageRejectsLegacyDurationPricingForNonVideoModel(t *testing.T) {

backend/internal/application/conversation/service_billing.go

@@ -278,6 +278,13 @@ func (s *Service) buildSendMessageUsageLedger(ctx context.Context, input SendMes
 	if result == nil {
 		return nil, nil
 	}
+	isVideoGeneration := sendMessageResultIsVideoGeneration(result)
+	mediaType := ""
+	inputImageCount := int64(0)
+	if isVideoGeneration {
+		mediaType = "video"
+		inputImageCount, _ = countAttachmentKinds(result.UserMessage.Attachments)
+	}
 	latencyMS := result.LatencyMS
 	if latencyMS <= 0 {
 		latencyMS = result.AssistantMessage.LatencyMS
@@ -305,8 +312,10 @@ func (s *Service) buildSendMessageUsageLedger(ctx context.Context, input SendMes
 		OutputTokens:        result.AssistantMessage.OutputTokens,
 		ReasoningTokens:     result.AssistantMessage.ReasoningTokens,
 		CallCount:           1,
-		DurationBillable:    sendMessageResultIsVideoGeneration(result),
+		DurationBillable:    isVideoGeneration,
 		DurationSeconds:     sendMessageBillingDurationSeconds(result),
+		MediaType:           mediaType,
+		InputImageCount:     inputImageCount,
 		LatencyMS:           latencyMS,
 		ServerSideToolUsage: result.ServerSideToolUsage,
 		RawUsageJSON:        result.RawUsageJSON,

frontend/features/admin/components/sections/logs/admin-logs.tsx

@@ -167,13 +167,20 @@ function formatUsageBalance(value: number | null | undefined, billingDisplay: Bi
   return value === null || value === undefined ? "-" : formatBillingBalance(value, billingDisplay);
 }
 
+function usageBillableOutputTokens(item: AdminUsageLogDTO): number {
+  return item.outputTokens + item.reasoningTokens;
+}
+
 function usageTotalTokens(item: AdminUsageLogDTO): number {
-  return item.inputTokens + item.cacheReadTokens + item.cacheWriteTokens + item.outputTokens + item.reasoningTokens;
+  return item.inputTokens + item.cacheReadTokens + item.cacheWriteTokens + usageBillableOutputTokens(item);
 }
 
 type UsagePricingSnapshot = {
   pricing_mode?: "token" | "call" | "duration" | "tiered" | string;
   provider_protocol?: string;
+  duration_billable?: boolean;
+  media_type?: string;
+  input_image_count?: number;
   cache_timeout?: string;
   fast_mode?: boolean;
   billing_speed?: string;
@@ -494,7 +501,7 @@ function buildUsageBillingTooltipLines(
   const outputRate = readUsageSnapshotNumber(snapshot, "output_nanousd_per_m_tokens");
   const cacheReadRate = readUsageSnapshotNumber(snapshot, "cache_read_nanousd_per_m_tokens");
   const cacheWriteRate = readUsageSnapshotNumber(snapshot, "cache_write_nanousd_per_m_tokens");
-  const billedOutputTokens = item.outputTokens + item.reasoningTokens;
+  const billedOutputTokens = usageBillableOutputTokens(item);
   const totalLine = usageTotalLine(item, labels, billingDisplay);
   const cacheWriteLabel = cacheWriteBillingLabel(snapshot, labels.billingDisplay);
   const cacheWriteNote = cacheWriteBillingNote(snapshot, labels.billingDisplay);
@@ -604,26 +611,84 @@ function UsageLogModelCell({ item, labels }: { item: AdminUsageLogDTO; labels: U
   );
 }
 
-function UsageLogTokenCell({ item, locale }: { item: AdminUsageLogDTO; locale: string }) {
+function UsageLogUsageCell({ item, locale }: { item: AdminUsageLogDTO; locale: string }) {
   const t = useTranslations("adminLogs.usage.tokens");
+  const snapshot = parseUsagePricingSnapshot(item.pricingSnapshotJSON);
+  const isVideoUsage = snapshot.media_type === "video" || snapshot.duration_billable === true;
+  if (isVideoUsage) {
+    const inputImageCount = typeof snapshot.input_image_count === "number" && Number.isFinite(snapshot.input_image_count) && snapshot.input_image_count >= 0
+      ? Math.trunc(snapshot.input_image_count)
+      : null;
+    const mediaUsage = [
+      { label: t("input"), value: inputImageCount === null ? "—" : t("imageCount", { count: inputImageCount }) },
+      { label: t("output"), value: t("secondCount", { count: item.durationSeconds }) },
+    ];
+    return (
+      <div className="grid min-w-[10.5rem] gap-1">
+        {mediaUsage.map((entry) => (
+          <span
+            key={entry.label}
+            className="inline-flex h-5 items-center justify-between gap-2 rounded-md bg-muted/45 px-1.5 text-[11px] leading-none text-muted-foreground"
+          >
+            <span>{entry.label}</span>
+            <span className="font-mono tabular-nums">{entry.value}</span>
+          </span>
+        ))}
+      </div>
+    );
+  }
   const tokens = [
     { label: t("inputShort"), value: item.inputTokens },
-    { label: t("outputShort"), value: item.outputTokens },
+    {
+      label: t("outputShort"),
+      value: usageBillableOutputTokens(item),
+      breakdown: {
+        visible: item.outputTokens,
+        reasoning: item.reasoningTokens,
+      },
+    },
     { label: t("cacheReadShort"), value: item.cacheReadTokens },
     { label: t("cacheWriteShort"), value: item.cacheWriteTokens },
   ];
 
   return (
     <div className="grid min-w-[10.5rem] grid-cols-2 gap-1">
-      {tokens.map((token) => (
-        <span
-          key={token.label}
-          className="inline-flex h-5 items-center justify-between gap-1 rounded-md bg-muted/45 px-1.5 font-mono text-[11px] leading-none text-muted-foreground"
-        >
-          <span>{token.label}</span>
-          <span className="tabular-nums">{formatCount(token.value, locale)}</span>
-        </span>
-      ))}
+      {tokens.map((token) => {
+        const badge = (
+          <span className={cn(
+            "inline-flex h-5 items-center justify-between gap-1 rounded-md bg-muted/45 px-1.5 font-mono text-[11px] leading-none text-muted-foreground",
+            token.breakdown && "cursor-help",
+          )}>
+            <span>{token.label}</span>
+            <span className="tabular-nums">{formatCount(token.value, locale)}</span>
+          </span>
+        );
+        if (!token.breakdown) {
+          return <React.Fragment key={token.label}>{badge}</React.Fragment>;
+        }
+        return (
+          <Tooltip key={token.label}>
+            <TooltipTrigger asChild>{badge}</TooltipTrigger>
+            <TooltipContent side="top" className="min-w-40">
+              <div className="grid gap-1.5 text-xs">
+                <div className="flex items-center justify-between gap-5">
+                  <span>{t("output")}</span>
+                  <span className="font-mono tabular-nums">{formatCount(token.breakdown.visible, locale)}</span>
+                </div>
+                <div className="flex items-center justify-between gap-5">
+                  <span>{t("reasoning")}</span>
+                  <span className="font-mono tabular-nums">{formatCount(token.breakdown.reasoning, locale)}</span>
+                </div>
+                <Separator className="bg-background/20" />
+                <div className="flex items-center justify-between gap-5 font-medium">
+                  <span>{t("outputTotal")}</span>
+                  <span className="font-mono tabular-nums">{formatCount(token.value, locale)}</span>
+                </div>
+              </div>
+            </TooltipContent>
+          </Tooltip>
+        );
+      })}
     </div>
   );
 }
@@ -1318,7 +1383,7 @@ function UsageLogTable({
             <TableHead className="w-[72px]">ID</TableHead>
             <TableHead>{t("columns.caller")}</TableHead>
             <TableHead>{t("columns.model")}</TableHead>
-            <TableHead>Token</TableHead>
+            <TableHead>{t("columns.usage")}</TableHead>
             <TableHead>{t("columns.billing")}</TableHead>
             <TableHead>{t("columns.balanceAfter")}</TableHead>
             <TableHead>{t("columns.latency")}</TableHead>
@@ -1340,7 +1405,7 @@ function UsageLogTable({
                 <UsageLogModelCell item={item} labels={usageLabels} />
               </TableCell>
               <TableCell>
-                <UsageLogTokenCell item={item} locale={locale} />
+                <UsageLogUsageCell item={item} locale={locale} />
               </TableCell>
               <TableCell><UsageLogCostCell item={item} labels={usageLabels} billingDisplay={billingDisplay} /></TableCell>
               <TableCell className="whitespace-nowrap font-medium tabular-nums text-foreground">

frontend/features/chat/components/message/message-bot.tsx

@@ -685,21 +685,12 @@ function MessageInlineVideoPreview({
   const resolveErrorMessage = useLocalizedErrorMessage();
   const objectURLRef = React.useRef<string | null>(null);
   const [state, setState] = React.useState<InlineVideoPreviewState>({ status: "loading" });
-  const [detectedDurationSeconds, setDetectedDurationSeconds] = React.useState<number | undefined>();
   const fileID = attachment.fileID;
   const fileName = attachment.fileName;
   const mimeType = attachment.mimeType;
   const detectedMime = attachment.detectedMime;
   const previewURL = attachment.previewURL;
   const sizeBytes = attachment.sizeBytes;
-  const displayDurationSeconds = attachment.durationSeconds ?? detectedDurationSeconds;
-
-  const handleDurationChange = React.useCallback((durationSeconds: number) => {
-    setDetectedDurationSeconds(
-      Number.isFinite(durationSeconds) && durationSeconds > 0 ? Math.ceil(durationSeconds) : undefined,
-    );
-  }, []);
-
   const revokeObjectURL = React.useCallback(() => {
     if (!objectURLRef.current) {
       return;
@@ -711,7 +702,6 @@ function MessageInlineVideoPreview({
   React.useEffect(() => {
     let cancelled = false;
     revokeObjectURL();
-    setDetectedDurationSeconds(undefined);
 
     if (previewURL) {
       setState({
@@ -795,23 +785,14 @@ function MessageInlineVideoPreview({
   }
 
   return (
-    <div className="relative my-4 w-full max-w-[40rem]">
+    <div className="my-4 w-full max-w-[40rem]">
       <PreviewMedia
         kind="video"
         source={state.source}
         alt={attachment.fileName}
         contentType={state.contentType}
         inline
-        onDurationChange={handleDurationChange}
       />
-      {displayDurationSeconds && displayDurationSeconds > 0 ? (
-        <span
-          className="pointer-events-none absolute left-3 top-3 rounded-md bg-black/65 px-2 py-1 text-xs font-medium tabular-nums text-white backdrop-blur-sm"
-          title={tMessages("videoDuration", { seconds: displayDurationSeconds })}
-        >
-          {displayDurationSeconds}s
-        </span>
-      ) : null}
     </div>
   );
 }

frontend/features/chat/hooks/use-chat-message-submit.ts

@@ -15,6 +15,7 @@ import type { ChatSubmitBlockReason } from "@/features/chat/model/chat-task";
 import { resolveChatSubmitDecision } from "@/features/chat/model/chat-task";
 import { useHiddenQueuedParentRuns } from "@/features/chat/hooks/use-hidden-queued-parent-runs";
 import {
+  resolveAssistantInputSideUsageValue,
   resolveDefaultSubmissionParentMessage,
   resolvePersistedPublicID,
   toPendingAttachments,
@@ -102,15 +103,6 @@ function streamEventErrorToApiError(
   return new ApiError(event.message || fallback, 502, event.debug, event.errorCode);
 }
 
-function resolveInputSideUsageValue(...values: Array<number | null | undefined>): number {
-  for (const value of values) {
-    if (typeof value === "number" && Number.isFinite(value) && value > 0) {
-      return value;
-    }
-  }
-  return 0;
-}
-
 function resolveMediaStatusLabel(
   status: string,
   fallbackMessage: string,
@@ -1195,18 +1187,21 @@ export function useChatMessageSubmit({
             assistantUpdatedAt: completed.assistantMessage.updatedAt,
             assistantContentType: completed.assistantMessage.contentType || current.assistantContentType,
             assistantAttachments: parseAttachments(completed.assistantMessage.attachments),
-            assistantInputTokens: resolveInputSideUsageValue(
+            assistantInputTokens: resolveAssistantInputSideUsageValue(
+              assistantOnlyBranch,
               completed.assistantMessage.inputTokens,
               completed.userMessage.inputTokens,
               current.assistantInputTokens,
             ),
             assistantOutputTokens: completed.assistantMessage.outputTokens,
-            assistantCacheReadTokens: resolveInputSideUsageValue(
+            assistantCacheReadTokens: resolveAssistantInputSideUsageValue(
+              assistantOnlyBranch,
               completed.assistantMessage.cacheReadTokens,
               completed.userMessage.cacheReadTokens,
               current.assistantCacheReadTokens,
             ),
-            assistantCacheWriteTokens: resolveInputSideUsageValue(
+            assistantCacheWriteTokens: resolveAssistantInputSideUsageValue(
+              assistantOnlyBranch,
               completed.assistantMessage.cacheWriteTokens,
               completed.userMessage.cacheWriteTokens,
               current.assistantCacheWriteTokens,

frontend/features/chat/model/chat-thread.ts

@@ -368,6 +368,12 @@ export function buildVisibleMessages(
     if (item.role !== "assistant") {
       return item;
     }
+    // Assistant-only retries reuse the original user message, but own the
+    // prompt-side usage for their generation. A zero value is authoritative
+    // and must not fall back to the reused user's first-run usage.
+    if (item.branchReason === "retry" && item.sourcePublicID?.trim()) {
+      return item;
+    }
     const previous = index > 0 ? withBranchNavigators[index - 1] : null;
     if (!previous || previous.role !== "user") {
       return item;

frontend/features/chat/model/message-submit.ts

@@ -41,6 +41,25 @@ export function resolvePersistedPublicID(value: string | null | undefined): stri
   return normalized;
 }
 
+export function resolveAssistantInputSideUsageValue(
+  assistantOwnsUsage: boolean,
+  assistantValue: number | null | undefined,
+  userValue: number | null | undefined,
+  liveValue: number | null | undefined,
+): number {
+  if (assistantOwnsUsage) {
+    return typeof assistantValue === "number" && Number.isFinite(assistantValue) && assistantValue >= 0
+      ? assistantValue
+      : 0;
+  }
+  for (const value of [assistantValue, userValue, liveValue]) {
+    if (typeof value === "number" && Number.isFinite(value) && value > 0) {
+      return value;
+    }
+  }
+  return 0;
+}
+
 function isSuccessfulContextMessage(message: ChatAreaMessage): boolean {
   const status = message.status?.trim().toLowerCase() || "success";
   return (

frontend/i18n/messages/en-US/admin-logs.json

@@ -104,6 +104,7 @@
     "message": "Message",
     "caller": "Caller",
     "model": "Model",
+    "usage": "Usage",
     "billing": "Billing",
     "balanceAfter": "Balance after settlement",
     "latency": "Latency",
@@ -203,7 +204,13 @@
       "inputShort": "In",
       "outputShort": "Out",
       "cacheReadShort": "Read",
-      "cacheWriteShort": "Write"
+      "cacheWriteShort": "Write",
+      "input": "Input",
+      "output": "Output",
+      "reasoning": "Reasoning",
+      "outputTotal": "Output total",
+      "imageCount": "{count, plural, one {# image} other {# images}}",
+      "secondCount": "{count, plural, one {# second} other {# seconds}}"
     },
     "billing": {
       "cacheWrite": "Cache write",

frontend/i18n/messages/en-US/chat.json

@@ -187,7 +187,6 @@
     "contextCompressed": "Context was automatically compressed",
     "scrollToBottom": "Scroll to bottom",
     "processing": "Processing...",
-    "videoDuration": "Video duration: {seconds} seconds",
     "expandUserMessage": "Expand",
     "collapseUserMessage": "Collapse",
     "editCreatesBranch": "Saving creates a new branch in this conversation.",