diff --git a/controller/auth_session.go b/controller/auth_session.go
index 0f1599c5..3921c779 100644
--- a/controller/auth_session.go
+++ b/controller/auth_session.go
@@ -2,9 +2,11 @@ package controller
 
 import (
 	"errors"
+	"fmt"
 	"net/http"
 	"strings"
 
+	"github.com/QuantumNous/new-api/logger"
 	"github.com/QuantumNous/new-api/middleware"
 	"github.com/QuantumNous/new-api/model"
 	"github.com/QuantumNous/new-api/service"
@@ -164,6 +166,12 @@ func writeAuthSessionError(c *gin.Context, err error) {
 	if errors.Is(err, gorm.ErrRecordNotFound) {
 		status, code = http.StatusUnauthorized, "AUTH_UNAUTHORIZED"
 	}
+	if status == http.StatusInternalServerError {
+		// The response body only carries the generic AUTH_INTERNAL_ERROR
+		// code; without this log the underlying Redis/database/session
+		// failure is indistinguishable from the client side.
+		logger.LogError(c.Request.Context(), fmt.Sprintf("auth session internal error (%s %s): %v", c.Request.Method, c.Request.URL.Path, err))
+	}
 	c.JSON(status, gin.H{"success": false, "code": code, "message": http.StatusText(status)})
 }
 
diff --git a/middleware/rate-limit.go b/middleware/rate-limit.go
index a6389161..350ee8ec 100644
--- a/middleware/rate-limit.go
+++ b/middleware/rate-limit.go
@@ -8,6 +8,7 @@ import (
 	"strconv"
 
 	"github.com/QuantumNous/new-api/common"
+	"github.com/QuantumNous/new-api/logger"
 	"github.com/gin-gonic/gin"
 )
 
@@ -106,33 +107,43 @@ func redisFixedWindowTake(ctx context.Context, key string, maxRequestNum int, du
 }
 
 func redisRateLimiter(c *gin.Context, maxRequestNum int, duration int64, mark string) {
-	allowed, _, _, err := redisFixedWindowTake(
+	allowed, _, ttlSeconds, err := redisFixedWindowTake(
 		c.Request.Context(),
 		redisIPRateLimitKey(mark, c.ClientIP()),
 		maxRequestNum,
 		duration,
 	)
 	if err != nil {
-		fmt.Println(err.Error())
+		logger.LogError(c.Request.Context(), fmt.Sprintf("rate limit check failed (mark=%s): %v", mark, err))
 		c.Status(http.StatusInternalServerError)
 		c.Abort()
 		return
 	}
 	if !allowed {
-		c.Status(http.StatusTooManyRequests)
-		c.Abort()
+		writeRateLimited(c, ttlSeconds)
 	}
 }
 
 func memoryRateLimiter(c *gin.Context, maxRequestNum int, duration int64, mark string) {
 	key := mark + c.ClientIP()
 	if !inMemoryRateLimiter.Request(key, maxRequestNum, duration) {
-		c.Status(http.StatusTooManyRequests)
-		c.Abort()
+		writeRateLimited(c, duration)
 		return
 	}
 }
 
+// writeRateLimited rejects the request with 429 and a Retry-After hint so
+// clients can back off instead of treating the rejection as a fatal error.
+// The in-memory limiter cannot report the remaining window, so callers
+// without a TTL pass the full window duration as a conservative upper bound.
+func writeRateLimited(c *gin.Context, retryAfterSeconds int64) {
+	if retryAfterSeconds > 0 {
+		c.Header("Retry-After", strconv.FormatInt(retryAfterSeconds, 10))
+	}
+	c.Status(http.StatusTooManyRequests)
+	c.Abort()
+}
+
 func rateLimitFactory(maxRequestNum int, duration int64, mark string) func(c *gin.Context) {
 	if common.RedisEnabled {
 		return func(c *gin.Context) {
@@ -201,8 +212,7 @@ func userRateLimitFactory(maxRequestNum int, duration int64, mark string) func(c
 		}
 		key := fmt.Sprintf("%s:user:%d", mark, userID)
 		if !inMemoryRateLimiter.Request(key, maxRequestNum, duration) {
-			c.Status(http.StatusTooManyRequests)
-			c.Abort()
+			writeRateLimited(c, duration)
 			return
 		}
 	}
@@ -211,16 +221,15 @@ func userRateLimitFactory(maxRequestNum int, duration int64, mark string) func(c
 // userRedisRateLimiter is like redisRateLimiter but accepts a pre-built key
 // (to support user-ID-based keys).
 func userRedisRateLimiter(c *gin.Context, maxRequestNum int, duration int64, key string) {
-	allowed, _, _, err := redisFixedWindowTake(c.Request.Context(), key, maxRequestNum, duration)
+	allowed, _, ttlSeconds, err := redisFixedWindowTake(c.Request.Context(), key, maxRequestNum, duration)
 	if err != nil {
-		fmt.Println(err.Error())
+		logger.LogError(c.Request.Context(), fmt.Sprintf("rate limit check failed (key=%s): %v", key, err))
 		c.Status(http.StatusInternalServerError)
 		c.Abort()
 		return
 	}
 	if !allowed {
-		c.Status(http.StatusTooManyRequests)
-		c.Abort()
+		writeRateLimited(c, ttlSeconds)
 	}
 }
 
diff --git a/middleware/rate_limit_test.go b/middleware/rate_limit_test.go
index b955753b..1ca48372 100644
--- a/middleware/rate_limit_test.go
+++ b/middleware/rate_limit_test.go
@@ -61,7 +61,9 @@ func TestRedisIPRateLimiterThresholdTTLAndNamespace(t *testing.T) {
 	require.NoError(t, err)
 	assert.Equal(t, http.StatusNoContent, performRateLimitRequest(router, "/limited", remoteAddr).Code)
 	assert.Equal(t, http.StatusNoContent, performRateLimitRequest(router, "/limited", remoteAddr).Code)
-	assert.Equal(t, http.StatusTooManyRequests, performRateLimitRequest(router, "/limited", remoteAddr).Code)
+	limitedResponse := performRateLimitRequest(router, "/limited", remoteAddr)
+	assert.Equal(t, http.StatusTooManyRequests, limitedResponse.Code)
+	assert.Equal(t, "37", limitedResponse.Header().Get("Retry-After"))
 
 	key := redisIPRateLimitKey("TEST", "192.0.2.10")
 	count, err := redisServer.Get(key)
diff --git a/web/src/lib/auth-session.test.ts b/web/src/lib/auth-session.test.ts
index 7797684b..4a97d3c2 100644
--- a/web/src/lib/auth-session.test.ts
+++ b/web/src/lib/auth-session.test.ts
@@ -147,6 +147,30 @@ describe('authentication session coordination', () => {
     assert.equal(transientCount, 1)
   })
 
+  test('a rate limited refresh remains retryable without clearing the session', async () => {
+    let transientCount = 0
+    let clearCount = 0
+    const runtime: AuthRefreshRuntime = {
+      request: async () => ({ status: 429 }),
+      getExpectedSID: () => bundle.session.sid,
+      parseBundle: () => null,
+      acceptBundle: () => undefined,
+      clear: () => {
+        clearCount += 1
+      },
+      markTransient: () => {
+        transientCount += 1
+      },
+      wait: async () => undefined,
+    }
+
+    const outcome = await createRefreshRunner(runtime)()
+
+    assert.equal(outcome.kind, 'transient_error')
+    assert.equal(clearCount, 0)
+    assert.equal(transientCount, 1)
+  })
+
   test('an exhausted refresh race clears the unusable local session', async () => {
     const requestedDelays: number[] = []
     const clears: Array<[boolean, string | undefined]> = []
diff --git a/web/src/lib/auth-session.ts b/web/src/lib/auth-session.ts
index 3921af6b..67c5fe8b 100644
--- a/web/src/lib/auth-session.ts
+++ b/web/src/lib/auth-session.ts
@@ -253,7 +253,7 @@ export function createRefreshRunner(
       return { kind: 'anonymous' }
     }
 
-    if (!response.status || response.status >= 500) {
+    if (!response.status || response.status >= 500 || response.status === 429) {
       runtime.markTransient()
       return {
         kind: 'transient_error',