diff --git a/.github/workflows/release-vllm-docker.yml b/.github/workflows/release-vllm-docker.yml
index 03bb32dde2..274c95a011 100644
--- a/.github/workflows/release-vllm-docker.yml
+++ b/.github/workflows/release-vllm-docker.yml
@@ -2,7 +2,7 @@ name: Release SMG+vLLM Docker Image
 
 run-name: >-
   SMG+vLLM |
-  ${{ github.event_name == 'push' && '3-version matrix' || format('base={0}', inputs.base_image_ref || 'vllm/vllm-openai:v0.22.1') }} |
+  ${{ github.event_name == 'push' && '3-version matrix' || format('base={0}', inputs.base_image_ref || 'vllm/vllm-openai:v0.26.0') }} |
   engine=${{ inputs.vllm_commit || 'latest' }} |
   smg=${{ github.event_name == 'pull_request' && 'main' || inputs.smg_commit || 'v1.8.0' }} |
   ${{ github.event_name == 'pull_request' && 'dry-run |' || '' }}
@@ -23,8 +23,8 @@ on:
   workflow_dispatch:
     inputs:
       base_image_ref:
-        description: 'Base image (e.g. vllm/vllm-openai:v0.22.1)'
-        default: 'vllm/vllm-openai:v0.22.1'
+        description: 'Base image (e.g. vllm/vllm-openai:v0.26.0)'
+        default: 'vllm/vllm-openai:v0.26.0'
         required: true
         type: string
       vllm_repo:
@@ -47,7 +47,7 @@ on:
         default: 'v1.8.0'
         type: string
       tag:
-        description: 'Override image tag (e.g. v1.8.0-vllm-v0.22.1)'
+        description: 'Override image tag (e.g. v1.8.0-vllm-v0.26.0)'
         required: false
         type: string
 
@@ -60,7 +60,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        base_image: ${{ github.event_name == 'push' && fromJSON('["vllm/vllm-openai:v0.22.1","vllm/vllm-openai:v0.21.0","vllm/vllm-openai:v0.20.0"]') || fromJSON(format('["{0}"]', inputs.base_image_ref || 'vllm/vllm-openai:v0.22.1')) }}
+        base_image: ${{ github.event_name == 'push' && fromJSON('["vllm/vllm-openai:v0.26.0","vllm/vllm-openai:v0.25.0","vllm/vllm-openai:v0.23.0"]') || fromJSON(format('["{0}"]', inputs.base_image_ref || 'vllm/vllm-openai:v0.26.0')) }}
     uses: ./.github/workflows/_build-engine-image.yml
     with:
       engine: vllm
diff --git a/scripts/ci_install_vllm.sh b/scripts/ci_install_vllm.sh
index 29c197372a..74afb532e1 100755
--- a/scripts/ci_install_vllm.sh
+++ b/scripts/ci_install_vllm.sh
@@ -19,14 +19,12 @@ fi
 
 echo "Using uv version: $(uv --version)"
 
-# Floor 0.25.0: older vllm releases do not guarantee torchcodec, while the
-# import canary below deliberately validates it. This line also admits only
-# transformers >= 5.5.3, preserving e5-mistral last-token pooling.
-# FastAPI 0.137 makes vLLM's prometheus-fastapi-instrumentator health route
-# crash on _IncludedRouter entries; keep the last known-good FastAPI line.
+# Floor 0.26.0: older vllm releases do not guarantee torchcodec, while the
+# import canary below deliberately validates it, and 0.26 caps
+# fastapi<0.137 (0.137 breaks the prometheus-fastapi-instrumentator health route).
 # --torch-backend=auto matches the torch CUDA variant to the pod's driver.
 echo "Installing vLLM..."
-uv pip install "vllm>=0.25.0" "fastapi<0.137" --torch-backend=auto
+uv pip install "vllm>=0.26.0" --torch-backend=auto
 
 # vLLM >=0.25 eagerly imports torchcodec, which dlopens the FFmpeg shared
 # libraries (libavutil/libavcodec/libavformat/...) at import time. The runner