Add Mistral Medium 3.1 and reduce Mistral retries

Author: Android-PowerUser

app/src/main/kotlin/com/google/ai/sample/GenerativeAiViewModelFactory.kt

@@ -31,6 +31,7 @@ enum class ModelOption(
 ) {
     PUTER_GLM5("GLM-5 (Puter)", "z-ai/glm-5", ApiProvider.PUTER, supportsScreenshot = false),
     MISTRAL_LARGE_3("Mistral Large 3", "mistral-large-latest", ApiProvider.MISTRAL),
+    MISTRAL_MEDIUM_3_1("Mistral Medium 3.1", "mistral-medium-latest", ApiProvider.MISTRAL),
     GPT_5_1_CODEX_MAX("GPT-5.1 Codex Max (Vercel)", "openai/gpt-5.1-codex-max", ApiProvider.VERCEL),
     GPT_5_1_CODEX_MINI("GPT-5.1 Codex Mini (Vercel)", "openai/gpt-5.1-codex-mini", ApiProvider.VERCEL),
     GPT_5_NANO("GPT-5 Nano (Vercel)", "openai/gpt-5-nano", ApiProvider.VERCEL),

app/src/main/kotlin/com/google/ai/sample/ScreenCaptureApiClients.kt

@@ -134,7 +134,20 @@ internal suspend fun callMistralApi(
             .build()
 
         val keysForCoordinator = availableApiKeys.filter { it.isNotBlank() }.distinct().ifEmpty { listOf(apiKey) }
-        val coordinated = MistralRequestCoordinator.execute(apiKeys = keysForCoordinator, maxAttempts = maxOf(4, keysForCoordinator.size * 3)) { key ->
+        val minIntervalMs = if (modelName == com.google.ai.sample.ModelOption.MISTRAL_MEDIUM_3_1.modelName) 420L else 1500L
+        val maxAttempts = if (
+            modelName == com.google.ai.sample.ModelOption.MISTRAL_LARGE_3.modelName ||
+            modelName == com.google.ai.sample.ModelOption.MISTRAL_MEDIUM_3_1.modelName
+        ) {
+            3
+        } else {
+            maxOf(4, keysForCoordinator.size * 3)
+        }
+        val coordinated = MistralRequestCoordinator.execute(
+            apiKeys = keysForCoordinator,
+            maxAttempts = maxAttempts,
+            minIntervalMs = minIntervalMs
+        ) { key ->
             client.newCall(
                 request.newBuilder()
                     .header("Authorization", "Bearer $key")

app/src/main/kotlin/com/google/ai/sample/feature/multimodal/PhotoReasoningViewModel.kt

@@ -1142,10 +1142,19 @@ class PhotoReasoningViewModel(
 
             // Validate that we have at least one key before proceeding
             require(availableKeys.isNotEmpty()) { "No valid Mistral API keys available after filtering" }
-            val maxAttempts = availableKeys.size * 4 + 8
+            val mistralMinIntervalMs = when (currentModel) {
+                ModelOption.MISTRAL_MEDIUM_3_1 -> 420L
+                else -> 1500L
+            }
+            val maxAttempts = when (currentModel) {
+                ModelOption.MISTRAL_LARGE_3,
+                ModelOption.MISTRAL_MEDIUM_3_1 -> 3
+                else -> availableKeys.size * 4 + 8
+            }
             val coordinated = MistralRequestCoordinator.execute(
                 apiKeys = availableKeys,
-                maxAttempts = maxAttempts
+                maxAttempts = maxAttempts,
+                minIntervalMs = mistralMinIntervalMs
             ) { selectedKey ->
                 if (stopExecutionFlag.get()) {
                     throw IOException("Mistral request aborted.")

app/src/main/kotlin/com/google/ai/sample/network/MistralRequestCoordinator.kt

@@ -30,9 +30,10 @@ internal object MistralRequestCoordinator {
     private suspend fun markKeyCooldown(
         key: String,
         referenceTimeMs: Long,
+        minIntervalMs: Long,
         extraDelayMs: Long = 0L
     ) {
-        val nextAllowedAt = referenceTimeMs + max(MIN_INTERVAL_MS, extraDelayMs.coerceAtLeast(0L))
+        val nextAllowedAt = referenceTimeMs + max(minIntervalMs.coerceAtLeast(0L), extraDelayMs.coerceAtLeast(0L))
         cooldownMutex.withLock {
             val existing = nextAllowedRequestAtMsByKey[key] ?: 0L
             nextAllowedRequestAtMsByKey[key] = max(existing, nextAllowedAt)
@@ -77,6 +78,7 @@ internal object MistralRequestCoordinator {
     suspend fun execute(
         apiKeys: List<String>,
         maxAttempts: Int = apiKeys.size * 4 + 8,
+        minIntervalMs: Long = MIN_INTERVAL_MS,
         request: suspend (apiKey: String) -> Response
     ): MistralCoordinatedResponse {
         require(apiKeys.isNotEmpty()) { "No Mistral API keys provided." }
@@ -120,7 +122,7 @@ internal object MistralRequestCoordinator {
                     TAG,
                     "[$rid] response code=${response.code}, retryAfterMs=${retryAfterMs ?: -1}, resetDelayMs=${resetDelayMs ?: -1}, appliedDelayMs=$serverRequestedDelayMs"
                 )
-                markKeyCooldown(selectedKey, requestEndMs, serverRequestedDelayMs)
+                markKeyCooldown(selectedKey, requestEndMs, minIntervalMs, serverRequestedDelayMs)
 
                 if (response.isSuccessful || !isRetryableFailure(response.code)) {
                     Log.d(TAG, "[$rid] returning response code=${response.code} with key=${keyFingerprint(selectedKey)}")
@@ -135,7 +137,7 @@ internal object MistralRequestCoordinator {
                     TAG,
                     "[$rid] retryable failure code=${response.code}, consecutiveFailures=$consecutiveFailures, adaptiveDelay=$adaptiveDelay"
                 )
-                markKeyCooldown(selectedKey, requestEndMs, max(serverRequestedDelayMs, adaptiveDelay))
+                markKeyCooldown(selectedKey, requestEndMs, minIntervalMs, max(serverRequestedDelayMs, adaptiveDelay))
             } catch (e: Exception) {
                 val requestEndMs = System.currentTimeMillis()
                 blockedKeysThisRound.add(selectedKey)
@@ -145,7 +147,7 @@ internal object MistralRequestCoordinator {
                     "[$rid] exception on key=${keyFingerprint(selectedKey)}, consecutiveFailures=$consecutiveFailures: ${e.message}",
                     e
                 )
-                markKeyCooldown(selectedKey, requestEndMs, adaptiveRetryDelayMs(consecutiveFailures))
+                markKeyCooldown(selectedKey, requestEndMs, minIntervalMs, adaptiveRetryDelayMs(consecutiveFailures))
                 if (consecutiveFailures >= maxAttempts) throw e
             }
         }