feat(statistics): show cache tokens and report whole-turn totals

app/src/main/java/com/zhousl/aether/data/ChatRepository.kt

@@ -1255,6 +1255,8 @@ private fun parseUsageStatistics(json: JSONObject?): ChatUsageStatistics? {
         totalTokens = json.optionalLong("totalTokens"),
         reasoningTokens = json.optionalLong("reasoningTokens"),
         cachedInputTokens = json.optionalLong("cachedInputTokens"),
+        cacheWriteTokens = json.optionalLong("cacheWriteTokens"),
+        outputDurationMillis = json.optionalLong("outputDurationMillis"),
         requestCount = json.optInt("requestCount", 1).coerceAtLeast(1),
         tokenUsageSource = json.optString("tokenUsageSource").ifBlank { "unavailable" },
         startedAtMillis = json.optLong("startedAtMillis"),
@@ -1269,6 +1271,8 @@ private fun ChatUsageStatistics.toJson(): JSONObject = JSONObject().apply {
     totalTokens?.let { put("totalTokens", it) }
     reasoningTokens?.let { put("reasoningTokens", it) }
     cachedInputTokens?.let { put("cachedInputTokens", it) }
+    cacheWriteTokens?.let { put("cacheWriteTokens", it) }
+    outputDurationMillis?.let { put("outputDurationMillis", it) }
     put("requestCount", requestCount)
     put("tokenUsageSource", tokenUsageSource)
     if (startedAtMillis > 0L) put("startedAtMillis", startedAtMillis)

app/src/main/java/com/zhousl/aether/data/LlmModels.kt

@@ -8,6 +8,8 @@ data class LlmTokenUsage(
     val totalTokens: Long? = null,
     val reasoningTokens: Long? = null,
     val cachedInputTokens: Long? = null,
+    val cacheWriteTokens: Long? = null,
+    val outputDurationMillis: Long? = null,
     val requestCount: Int = 1,
 ) {
     operator fun plus(other: LlmTokenUsage): LlmTokenUsage = LlmTokenUsage(
@@ -16,6 +18,8 @@ data class LlmTokenUsage(
         totalTokens = sumNullable(totalTokens, other.totalTokens),
         reasoningTokens = sumNullable(reasoningTokens, other.reasoningTokens),
         cachedInputTokens = sumNullable(cachedInputTokens, other.cachedInputTokens),
+        cacheWriteTokens = sumNullable(cacheWriteTokens, other.cacheWriteTokens),
+        outputDurationMillis = sumNullable(outputDurationMillis, other.outputDurationMillis),
         requestCount = requestCount + other.requestCount,
     )
 

app/src/main/java/com/zhousl/aether/data/SessionExecutionManager.kt

@@ -1465,6 +1465,8 @@ class SessionExecutionManager(
             totalTokens = tokenUsage?.withMissingTotalResolved()?.totalTokens,
             reasoningTokens = tokenUsage?.reasoningTokens,
             cachedInputTokens = tokenUsage?.cachedInputTokens,
+            cacheWriteTokens = tokenUsage?.cacheWriteTokens,
+            outputDurationMillis = tokenUsage?.outputDurationMillis,
             requestCount = tokenUsage?.requestCount ?: 1,
             tokenUsageSource = tokenUsageSource,
             startedAtMillis = turnStartedAtMillis ?: 0L,

app/src/main/java/com/zhousl/aether/data/pi/PiProviderMapper.kt

@@ -232,6 +232,8 @@ fun PiCompletionResult.toProviderPayloadJson(): String = JSONObject().apply {
                 tokenUsage.totalTokens?.let { put("total_tokens", it) }
                 tokenUsage.reasoningTokens?.let { put("reasoning_tokens", it) }
                 tokenUsage.cachedInputTokens?.let { put("cached_input_tokens", it) }
+                tokenUsage.cacheWriteTokens?.let { put("cache_write_tokens", it) }
+                tokenUsage.outputDurationMillis?.let { put("output_duration_ms", it) }
                 put("request_count", tokenUsage.requestCount)
             },
         )
@@ -266,13 +268,17 @@ private fun JSONObject.toLlmTokenUsage(): LlmTokenUsage? {
         totalTokens = optPositiveLong("total_tokens"),
         reasoningTokens = optPositiveLong("reasoning_tokens"),
         cachedInputTokens = optPositiveLong("cached_input_tokens"),
+        cacheWriteTokens = optPositiveLong("cache_write_tokens"),
+        outputDurationMillis = optPositiveLong("output_duration_ms"),
+        requestCount = (optPositiveLong("request_count") ?: 1L).coerceAtLeast(1L).toInt(),
     ).withMissingTotalResolved()
     return usage.takeIf {
         it.inputTokens != null ||
             it.outputTokens != null ||
             it.totalTokens != null ||
             it.reasoningTokens != null ||
-            it.cachedInputTokens != null
+            it.cachedInputTokens != null ||
+            it.cacheWriteTokens != null
     }
 }
 

app/src/main/java/com/zhousl/aether/ui/AetherUiState.kt

@@ -111,6 +111,8 @@ data class ChatUsageStatistics(
     val totalTokens: Long? = null,
     val reasoningTokens: Long? = null,
     val cachedInputTokens: Long? = null,
+    val cacheWriteTokens: Long? = null,
+    val outputDurationMillis: Long? = null,
     val requestCount: Int = 1,
     val tokenUsageSource: String = "unavailable",
     val startedAtMillis: Long = 0L,
@@ -125,6 +127,10 @@ data class ChatUsageStatistics(
     val outputTokensPerSecond: Double?
         get() {
             val output = outputTokens ?: return null
+            val outputDuration = outputDurationMillis?.takeIf { it > 0L }
+            if (outputDuration != null) {
+                return output * 1_000.0 / outputDuration
+            }
             val outputStartedAt = firstTokenAtMillis ?: startedAtMillis.takeIf { it > 0L } ?: return null
             if (completedAtMillis <= outputStartedAt) return null
             val seconds = (completedAtMillis - outputStartedAt) / 1000.0

app/src/main/java/com/zhousl/aether/ui/AetherViewModel.kt

@@ -5058,6 +5058,7 @@ class AetherViewModel(
             put("total_tokens", totalTokens)
             usage?.reasoningTokens?.let { put("reasoning_tokens", it) }
             usage?.cachedInputTokens?.let { put("cached_input_tokens", it) }
+            usage?.cacheWriteTokens?.let { put("cache_write_tokens", it) }
             put(
                 "average_tokens_per_input_message",
                 if (inputMessageCount > 0) totalTokens.toDouble() / inputMessageCount else 0.0,

app/src/main/java/com/zhousl/aether/ui/ConversationMessages.kt

@@ -3413,6 +3413,16 @@ private fun MessageStatisticsPopup(
                 value = usageStatistics?.reasoningTokens?.let(::formatTokenCount)
                     ?: stringResource(R.string.statistics_unavailable),
             )
+            MessageStatisticRow(
+                label = stringResource(R.string.statistics_cached_input),
+                value = usageStatistics?.cachedInputTokens?.let(::formatTokenCount)
+                    ?: stringResource(R.string.statistics_unavailable),
+            )
+            MessageStatisticRow(
+                label = stringResource(R.string.statistics_cache_write),
+                value = usageStatistics?.cacheWriteTokens?.let(::formatTokenCount)
+                    ?: stringResource(R.string.statistics_unavailable),
+            )
             MessageStatisticRow(
                 label = stringResource(R.string.statistics_output_rate),
                 value = usageStatistics?.outputTokensPerSecond?.let(::formatTokenRate)

app/src/main/java/com/zhousl/aether/ui/SettingsScreen.kt

@@ -1704,6 +1704,29 @@ private fun StatisticsSettingsPage(
 
         Spacer(Modifier.height(16.dp))
 
+        SettingsCardGroup {
+            Column(
+                modifier = Modifier.padding(16.dp),
+                verticalArrangement = Arrangement.spacedBy(14.dp),
+            ) {
+                Text(
+                    text = stringResource(R.string.statistics_cache_tokens),
+                    style = MaterialTheme.typography.titleMedium,
+                    color = AetherOnSurface,
+                )
+                HistoryPeakRow(
+                    label = stringResource(R.string.statistics_cached_input),
+                    value = formatSettingsTokenCount(report.cachedInputTokens),
+                )
+                HistoryPeakRow(
+                    label = stringResource(R.string.statistics_cache_write),
+                    value = formatSettingsTokenCount(report.cacheWriteTokens),
+                )
+            }
+        }
+
+        Spacer(Modifier.height(16.dp))
+
         SettingsCardGroup {
             StatisticsChartSection(
                 title = stringResource(R.string.statistics_speed),
@@ -2130,6 +2153,8 @@ private fun buildUsageStatisticsReport(
         inputTokens = stats.sumOf { it.inputTokens ?: 0L },
         outputTokens = stats.sumOf { it.outputTokens ?: 0L },
         reasoningTokens = stats.sumOf { it.reasoningTokens ?: 0L },
+        cachedInputTokens = stats.sumOf { it.cachedInputTokens ?: 0L },
+        cacheWriteTokens = stats.sumOf { it.cacheWriteTokens ?: 0L },
         sessionCount = usageStatisticsSnapshots.map { it.sessionId }.distinct().size,
         turnCount = stats.size,
         dailyTokenUsage = daily,
@@ -2162,6 +2187,8 @@ private data class UsageStatisticsReport(
     val inputTokens: Long,
     val outputTokens: Long,
     val reasoningTokens: Long,
+    val cachedInputTokens: Long,
+    val cacheWriteTokens: Long,
     val sessionCount: Int,
     val turnCount: Int,
     val dailyTokenUsage: List<DailyTokenUsage>,

app/src/test/java/com/zhousl/aether/data/pi/PiProviderMapperTest.kt

@@ -251,7 +251,14 @@ class PiProviderMapperTest {
         val payload = PiCompletionResult(
             assistantText = "done",
             assistantMessage = assistantMessage,
-            usage = LlmTokenUsage(inputTokens = 3, outputTokens = 2, totalTokens = 5),
+            usage = LlmTokenUsage(
+                inputTokens = 3,
+                outputTokens = 2,
+                totalTokens = 5,
+                cachedInputTokens = 1,
+                cacheWriteTokens = 4,
+                outputDurationMillis = 2_500,
+            ),
             provider = "openai",
             model = "gpt-5.4",
             responseId = "resp-1",
@@ -263,6 +270,40 @@ class PiProviderMapperTest {
         assertEquals("openai", wrapped.getString("provider"))
         assertEquals("resp-1", wrapped.getString("responseId"))
         assertEquals(5L, wrapped.getJSONObject("usage").getLong("total_tokens"))
+        assertEquals(4L, wrapped.getJSONObject("usage").getLong("cache_write_tokens"))
+        assertEquals(2_500L, wrapped.getJSONObject("usage").getLong("output_duration_ms"))
+    }
+
+    @Test
+    fun piUsageKeepsCacheReadAndWriteTokens() {
+        val usage = JSONObject()
+            .put(
+                "usage",
+                JSONObject()
+                    .put("input_tokens", 10)
+                    .put("output_tokens", 4)
+                    .put("total_tokens", 14)
+                    .put("cached_input_tokens", 6)
+                    .put("cache_write_tokens", 2)
+                    .put("output_duration_ms", 2_500)
+                    .put("request_count", 3),
+            )
+            .toPiCompletionResult()
+            .usage!!
+
+        assertEquals(6L, usage.cachedInputTokens)
+        assertEquals(2L, usage.cacheWriteTokens)
+        assertEquals(2_500L, usage.outputDurationMillis)
+        assertEquals(3, usage.requestCount)
+
+        val replayed = JSONObject(
+            PiCompletionResult(assistantText = "", usage = usage).toProviderPayloadJson(),
+        )
+            .getJSONObject("usage")
+        assertEquals(6L, replayed.getLong("cached_input_tokens"))
+        assertEquals(2L, replayed.getLong("cache_write_tokens"))
+        assertEquals(2_500L, replayed.getLong("output_duration_ms"))
+        assertEquals(3, replayed.getInt("request_count"))
     }
 
     @Test

iosApp/App/Settings/NativeSettingsView.swift

@@ -31,6 +31,9 @@ private let nativeSettingsPersianTranslations: [String: String] = [
     "Back up or restore settings, providers, skills, and conversations.": "از تنظیمات، ارائه‌دهندگان، مهارت‌ها و گفتگوها پشتیبان بگیرید یا آن‌ها را بازیابی کنید.",
     "Base URL": "نشانی پایه",
     "Browser login": "ورود با مرورگر",
+    "Cache tokens": "توکن‌های کش",
+    "Cached read": "خواندن از کش",
+    "Cached write": "نوشتن در کش",
     "Cancel": "لغو",
     "Check your connection and try again.": "اتصال خود را بررسی کرده و دوباره تلاش کنید.",
     "Choose Folder": "انتخاب پوشه",
@@ -2197,6 +2200,10 @@ private struct NativeStatisticsView: View {
                 tokenLine(model.text("Output", "输出"), value: long("outputTokens"), color: .green)
                 tokenLine(model.text("Reasoning", "推理"), value: long("reasoningTokens"), color: .orange)
             }
+            Section(model.text("Cache tokens", "缓存 token")) {
+                LabeledContent(model.text("Cached read", "缓存读取"), value: optionalTokens("cachedInputTokens"))
+                LabeledContent(model.text("Cached write", "缓存写入"), value: optionalTokens("cacheWriteTokens"))
+            }
             Section(model.text("History", "历史")) {
                 LabeledContent(model.text("Peak day", "峰值日期"), value: peakDay)
                 LabeledContent(model.text("Largest turn", "最大单轮"), value: optionalTokens("largestTurnTokens"))