diff --git a/packages/pi-coding-agent/src/core/auth-storage.ts b/packages/pi-coding-agent/src/core/auth-storage.ts
index 21f78fa42..981aaec7d 100644
--- a/packages/pi-coding-agent/src/core/auth-storage.ts
+++ b/packages/pi-coding-agent/src/core/auth-storage.ts
@@ -201,7 +201,7 @@ const BACKOFF_QUOTA_EXHAUSTED_MS = 30 * 60_000; // 30min for quota exhausted
 const BACKOFF_SERVER_ERROR_MS = 20_000; // 20s for 5xx server errors
 const BACKOFF_DEFAULT_MS = 60_000; // 60s fallback
 
-export type UsageLimitErrorType = "rate_limit" | "quota_exhausted" | "server_error" | "unknown";
+export type UsageLimitErrorType = "rate_limit" | "quota_exhausted" | "server_error" | "auth_error" | "unknown";
 
 /**
  * Get backoff duration for an error type.
diff --git a/packages/pi-coding-agent/src/core/retry-handler.ts b/packages/pi-coding-agent/src/core/retry-handler.ts
index 771223591..4f61902f9 100644
--- a/packages/pi-coding-agent/src/core/retry-handler.ts
+++ b/packages/pi-coding-agent/src/core/retry-handler.ts
@@ -197,8 +197,9 @@ export class RetryHandler {
 			}
 
 			// Cross-provider fallback — for rate limits with all creds backed off,
-			// or quota errors (which skip credential backoff entirely).
-			if (isRateLimit || isQuotaError) {
+			// quota errors, or auth errors (invalid/expired key — no point retrying).
+			const isAuthError = errorType === "auth_error";
+			if (isRateLimit || isQuotaError || isAuthError) {
 				const fallbackResult = await this._deps.fallbackResolver.findFallback(
 					this._deps.getModel()!,
 					errorType,
@@ -422,6 +423,7 @@ export class RetryHandler {
 		if (/quota|billing|exceeded.*limit|usage.*limit/i.test(err)) return "quota_exhausted";
 		if (/rate.?limit|too many requests|429|529|overloaded/i.test(err)) return "rate_limit";
 		if (/500|502|503|504|server.?error|internal.?error|service.?unavailable/i.test(err)) return "server_error";
+		if (/401|authentication.*error|invalid.*api.?key|api.?key.*invalid|api.?key.*expired|failed to authenticate|unauthorized/i.test(err)) return "auth_error";
 		return "unknown";
 	}
 
diff --git a/src/resources/extensions/search-the-web/native-search.ts b/src/resources/extensions/search-the-web/native-search.ts
index 709065306..65a772bf1 100644
--- a/src/resources/extensions/search-the-web/native-search.ts
+++ b/src/resources/extensions/search-the-web/native-search.ts
@@ -96,7 +96,7 @@ export function registerNativeSearchHooks(pi: NativeSearchPI): { getIsAnthropic:
     const wasAnthropic = isAnthropicProvider;
     isAnthropicProvider = event.model.provider === "anthropic";
 
-    const hasBrave = !!process.env.BRAVE_API_KEY;
+    const hasSearchKey = !!(process.env.BRAVE_API_KEY || process.env.TAVILY_API_KEY || process.env.OLLAMA_API_KEY);
 
     // When Anthropic (and not preferring Brave): disable custom search tools —
     // native web_search is server-side and more reliable.
@@ -121,9 +121,9 @@ export function registerNativeSearchHooks(pi: NativeSearchPI): { getIsAnthropic:
       ctx.ui.notify("Native Anthropic web search active", "info");
     } else if (isAnthropicProvider && preferBraveSearch() && !wasAnthropic && event.source !== "restore") {
       ctx.ui.notify("Brave search active (PREFER_BRAVE_SEARCH)", "info");
-    } else if (!isAnthropicProvider && !hasBrave) {
+    } else if (!isAnthropicProvider && !hasSearchKey) {
       ctx.ui.notify(
-        "Web search: Set BRAVE_API_KEY or use an Anthropic model for built-in search",
+        "Web search: Set BRAVE_API_KEY, TAVILY_API_KEY, or OLLAMA_API_KEY, or use an Anthropic model for built-in search",
         "warning"
       );
     }
diff --git a/src/resources/extensions/sf/learning/data/model-benchmarks.json b/src/resources/extensions/sf/learning/data/model-benchmarks.json
index 65130b963..b249642ed 100644
--- a/src/resources/extensions/sf/learning/data/model-benchmarks.json
+++ b/src/resources/extensions/sf/learning/data/model-benchmarks.json
@@ -231,6 +231,25 @@
     "context_window": 204800,
     "max_output_tokens": 131072
   },
+  "MiniMax-M2.7-highspeed": {
+    "swe_bench": null,
+    "swe_bench_verified": null,
+    "live_code_bench": null,
+    "human_eval": null,
+    "hle": null,
+    "aime_2026": null,
+    "gpqa": null,
+    "mmlu_pro": null,
+    "bbh": null,
+    "browse_comp": null,
+    "simple_qa": null,
+    "long_context_ruler": 95,
+    "arena_elo": null,
+    "instruction_following": null,
+    "source": "MiniMax M2.7-highspeed — fast tier of M2.7, same context/output limits, RULER ~95 inherited",
+    "context_window": 131072,
+    "max_output_tokens": 131072
+  },
   "MiniMax-M2.5": {
     "swe_bench": null,
     "swe_bench_verified": null,