diff --git a/.cache/vllm_serve_flags.json b/.cache/vllm_serve_flags.json
index cc7e5160..6a85d96c 100644
--- a/.cache/vllm_serve_flags.json
+++ b/.cache/vllm_serve_flags.json
@@ -18,7 +18,6 @@
     "--disable-uvicorn-access-log",
     "--enable-auto-tool-choice",
     "--enable-batch-sharded-sampling",
-    "--enable-bf16x3-router-gemm",
     "--enable-chunked-prefill",
     "--enable-cumem-allocator",
     "--enable-dbo",
@@ -37,10 +36,12 @@
     "--enable-logging-iteration-details",
     "--enable-lora",
     "--enable-mamba-cache-stochastic-rounding",
+    "--enable-mamba-shared-prefix-checkpoint",
     "--enable-mfu-metrics",
     "--enable-mixed-moe-lora-format",
     "--enable-mm-embeds",
     "--enable-moe-shared-loras",
+    "--enable-nccl-comm-suspend",
     "--enable-offline-docs",
     "--enable-per-request-metrics",
     "--enable-prefix-caching",
@@ -48,6 +49,7 @@
     "--enable-prompt-tokens-details",
     "--enable-request-id-headers",
     "--enable-return-routed-experts",
+    "--enable-scale-out",
     "--enable-server-load-tracking",
     "--enable-sleep-mode",
     "--enable-ssl-refresh",
@@ -64,6 +66,7 @@
     "--kv-sharing-fast-prefill",
     "--language-model-only",
     "--log-error-stack",
+    "--long-prefill-token-threshold-adaptive",
     "--mm-device-do-normalize",
     "--mm-encoder-only",
     "--numa-bind",
@@ -74,9 +77,11 @@
     "--skip-mm-profiling",
     "--skip-tokenizer-init",
     "--specialize-active-lora",
+    "--swa-bounded-replay",
     "--tokens-only",
     "--trust-remote-code",
     "--trust-request-chat-template",
+    "--trust-request-mm-kwargs",
     "--use-fp64-gumbel",
     "--use-replayssm",
     "--use-tqdm-on-load"
@@ -96,10 +101,12 @@
     "--allowed-media-domains",
     "--allowed-methods",
     "--allowed-origins",
+    "--anthropic-disabled-thinking-effort",
     "--api-key",
     "--api-server-count",
     "--attention-backend",
     "--attention-config",
+    "--aux-output-config",
     "--block-size",
     "--chat-template",
     "--chat-template-content-format",
@@ -131,6 +138,7 @@
     "--default-chat-template-kwargs",
     "--default-mm-loras",
     "--device-ids",
+    "--device-memory-utilization",
     "--diffusion-config",
     "--disable-access-log-for-endpoints",
     "--distributed-executor-backend",
@@ -139,9 +147,12 @@
     "--dp-supervisor-probe-failure-threshold",
     "--dp-supervisor-probe-interval-s",
     "--dp-supervisor-probe-timeout-s",
+    "--dp-sync-interval",
     "--dtype",
     "--ec-manager-config",
     "--ec-transfer-config",
+    "--elastic-ep-max-dp-size",
+    "--engram-config",
     "--eplb-config",
     "--expert-placement-strategy",
     "--fault-tolerance-config",
@@ -160,6 +171,7 @@
     "--io-processor-plugin",
     "--ir-op-priority",
     "--jit-monitor-mode",
+    "--kda-decode-backend",
     "--kda-prefill-backend",
     "--kernel-config",
     "--kv-cache-dtype",
@@ -174,6 +186,8 @@
     "--linear-backend",
     "--load-format",
     "--log-config-file",
+    "--log-level",
+    "--logging-config",
     "--logits-processors",
     "--logprobs-mode",
     "--long-prefill-token-threshold",
@@ -193,9 +207,11 @@
     "--max-cudagraph-capture-size",
     "--max-log-len",
     "--max-logprobs",
+    "--max-lora-cls-labels",
     "--max-lora-rank",
     "--max-loras",
     "--max-model-len",
+    "--max-num-active-seqs",
     "--max-num-batched-tokens",
     "--max-num-queued-reqs",
     "--max-num-queued-tokens",
@@ -268,6 +284,8 @@
     "--served-model-name",
     "--show-hidden-metrics-for-version",
     "--shutdown-timeout",
+    "--sleep-preserve-parameter-names",
+    "--sparse-indexer-topk-backend",
     "--spec-method",
     "--spec-model",
     "--spec-tokens",
@@ -287,15 +305,17 @@
     "--tool-call-parser",
     "--tool-parser-plugin",
     "--tool-server",
+    "--tool-strict-level",
     "--ubatch-size",
     "--uds",
     "--uvicorn-log-level",
     "--video-pruning-method",
     "--video-pruning-rate",
     "--watermark",
+    "--watermark-config",
     "--weight-transfer-config",
     "--worker-cls",
     "--worker-extension-cls"
   ],
-  "vllm_version": "0.29.0"
+  "vllm_version": "0.31.0"
 }
diff --git a/Dockerfile b/Dockerfile
index 9b31b568..8e984633 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -1,7 +1,7 @@
 # Worker image = official vLLM OpenAI server image + RunPod serverless wrapper.
 # vLLM upgrades are now a single build ARG:
 #   docker buildx build --build-arg VLLM_VERSION=v0.23.0 ...
-ARG VLLM_VERSION=v0.30.0
+ARG VLLM_VERSION=v0.31.0
 FROM vllm/vllm-openai:${VLLM_VERSION}
 # Re-declare so the stage can reference it in RUN steps below.
 ARG VLLM_VERSION