diff --git a/deploy/kubernetes/easyai-ai-gateway-cluster-release b/deploy/kubernetes/easyai-ai-gateway-cluster-release
index 5e61198..327267f 100755
--- a/deploy/kubernetes/easyai-ai-gateway-cluster-release
+++ b/deploy/kubernetes/easyai-ai-gateway-cluster-release
@@ -19,7 +19,7 @@ source "$config_file"
 : "${AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT:=24}"
 : "${AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT:=48}"
 : "${AI_GATEWAY_DATABASE_MAX_CONNS:=32}"
-: "${AI_GATEWAY_API_DATABASE_MAX_CONNS:=48}"
+: "${AI_GATEWAY_API_DATABASE_MAX_CONNS:=64}"
 : "${AI_GATEWAY_DATABASE_MIN_IDLE_CONNS:=4}"
 : "${AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY:=24}"
 : "${AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY:=24}"
diff --git a/deploy/kubernetes/easyai-ai-gateway-cluster-release.conf.example b/deploy/kubernetes/easyai-ai-gateway-cluster-release.conf.example
index d7322b4..88cdeff 100644
--- a/deploy/kubernetes/easyai-ai-gateway-cluster-release.conf.example
+++ b/deploy/kubernetes/easyai-ai-gateway-cluster-release.conf.example
@@ -10,7 +10,7 @@ AI_GATEWAY_WORKER_REPLICAS_HONGKONG=1
 AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT=24
 AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=48
 AI_GATEWAY_DATABASE_MAX_CONNS=32
-AI_GATEWAY_API_DATABASE_MAX_CONNS=48
+AI_GATEWAY_API_DATABASE_MAX_CONNS=64
 AI_GATEWAY_DATABASE_MIN_IDLE_CONNS=4
 AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY=24
 AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY=24
diff --git a/deploy/kubernetes/production/application.yaml b/deploy/kubernetes/production/application.yaml
index 38ef184..fa52b91 100644
--- a/deploy/kubernetes/production/application.yaml
+++ b/deploy/kubernetes/production/application.yaml
@@ -53,7 +53,7 @@ spec:
             - name: AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED
               value: "false"
             - name: AI_GATEWAY_DATABASE_MAX_CONNS
-              value: "48"
+              value: "64"
             - name: AI_GATEWAY_DATABASE_MIN_IDLE_CONNS
               value: "4"
             - name: AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY
@@ -184,7 +184,7 @@ spec:
             - name: AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED
               value: "false"
             - name: AI_GATEWAY_DATABASE_MAX_CONNS
-              value: "48"
+              value: "64"
             - name: AI_GATEWAY_DATABASE_MIN_IDLE_CONNS
               value: "4"
             - name: AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY
diff --git a/docs/operations/production-acceptance.md b/docs/operations/production-acceptance.md
index a53ee18..8db3b73 100644
--- a/docs/operations/production-acceptance.md
+++ b/docs/operations/production-acceptance.md
@@ -80,7 +80,8 @@ AI_GATEWAY_WORKER_REPLICAS_HONGKONG
 | P28 | 28 | 36 | 28 | 56 |
 | P32 | 32 | 40 | 32 | 64 |
 
-API Pod 使用独立发布配置 `AI_GATEWAY_API_DATABASE_MAX_CONNS`，生产同构验收固定为 `48`。
+API Pod 使用独立发布配置 `AI_GATEWAY_API_DATABASE_MAX_CONNS`，生产同构验收默认使用 `64`；
+也可通过 `AI_GATEWAY_ACCEPTANCE_API_DATABASE_MAX_CONNS` 覆盖本次验收值，无需修改 Go 代码。
 所有 API/Worker Pool 的 `AI_GATEWAY_DATABASE_MIN_IDLE_CONNS` 固定为 `4`，仅预热必要连接，
 其余连接按负载弹性创建，避免四个 Pod 在空闲时就占满 PostgreSQL 连接预算。各档表中的
 数据库池仍表示单个 Worker 的最大连接数。
diff --git a/scripts/cluster/run-production-acceptance.sh b/scripts/cluster/run-production-acceptance.sh
index 93edb13..c9c0d41 100755
--- a/scripts/cluster/run-production-acceptance.sh
+++ b/scripts/cluster/run-production-acceptance.sh
@@ -51,9 +51,15 @@ require_commands curl git jq node openssl sed
 : "${AI_GATEWAY_ACCEPTANCE_USER_ID:?}"
 : "${AI_GATEWAY_ACCEPTANCE_GEMINI_MODEL:=}"
 : "${AI_GATEWAY_ACCEPTANCE_VIDEO_MODEL:=}"
+: "${AI_GATEWAY_ACCEPTANCE_API_DATABASE_MAX_CONNS:=64}"
 : "${AI_GATEWAY_ACCEPTANCE_GATEWAY_TLS_SERVER_NAME:=}"
 : "${AI_GATEWAY_ACCEPTANCE_GATEWAYS:?}"
 : "${AI_GATEWAY_ACCEPTANCE_REAL_IMAGE_URLS:?}"
+[[ $AI_GATEWAY_ACCEPTANCE_API_DATABASE_MAX_CONNS =~ ^[1-9][0-9]*$ &&
+   $AI_GATEWAY_ACCEPTANCE_API_DATABASE_MAX_CONNS -le 256 ]] || {
+  echo 'AI_GATEWAY_ACCEPTANCE_API_DATABASE_MAX_CONNS must be between 1 and 256' >&2
+  exit 1
+}
 
 node "$cluster_root/scripts/release-manifest.mjs" validate "$release_manifest" >/dev/null
 release_sha=$(node "$cluster_root/scripts/release-manifest.mjs" get "$release_manifest" sourceSha)
@@ -667,7 +673,7 @@ mark_run_failed() {
 apply_capacity_profile() {
   local profile=$1
   local slots pool media global
-  local api_pool=48
+  local api_pool=$AI_GATEWAY_ACCEPTANCE_API_DATABASE_MAX_CONNS
   local min_idle=4
   case $profile in
     P24) slots=24; pool=32; media=24; global=48 ;;
@@ -899,7 +905,7 @@ wait_for_memory_recovery() {
 
 verify_runtime_gates() {
   local expected_slots expected_global expected_pool
-  local expected_api_pool=48
+  local expected_api_pool=$AI_GATEWAY_ACCEPTANCE_API_DATABASE_MAX_CONNS
   case $active_profile in
     P24) expected_slots=24; expected_pool=32; expected_global=48 ;;
     P28) expected_slots=28; expected_pool=36; expected_global=56 ;;