diff --git a/docs/source/compute/show-gpus-h100-8.txt b/docs/source/compute/show-gpus-h100-8.txt
index 17cf51a672f..e5e91059cf4 100644
--- a/docs/source/compute/show-gpus-h100-8.txt
+++ b/docs/source/compute/show-gpus-h100-8.txt
@@ -9,7 +9,7 @@ H100  8    GCP         a3-highgpu-8g                     80GB        208    1872
 H100  8    Paperspace  H100x8                            -           128    640GB     $ 47.600      -                  East Coast (NY2)   
 H100  8    DO          gpu-h100x8-640gb                  80GB        160    1920GB    $ 47.600      -                  tor1               
 H100  8    OCI         BM.GPU.H100.8                     80GB        224    2048GB    $ 80.000      -                  eu-amsterdam-1     
-H100  8    AWS         p5.48xlarge                       80GB        192    2048GB    $ 98.320      $ 13.127           us-east-2          
+H100  8    AWS         p5.48xlarge                       80GB        192    2048GB    $ 98.320      $ 9.832            us-east-2          
 
 GPU        QTY  CLOUD  INSTANCE_TYPE  DEVICE_MEM  vCPUs  HOST_MEM  HOURLY_PRICE  HOURLY_SPOT_PRICE  REGION       
 H100-MEGA  8    GCP    a3-megagpu-8g  80GB        208    1872GB    $ 92.214      $ 21.208           us-central1  
diff --git a/docs/source/examples/models/index.rst b/docs/source/examples/models/index.rst
index 6fe280ce2fb..a1bdd18a2bf 100644
--- a/docs/source/examples/models/index.rst
+++ b/docs/source/examples/models/index.rst
@@ -8,6 +8,7 @@ Models
    DeepSeek-R1 Distilled <deepseek-r1-distilled>
    DeepSeek-Janus <deepseek-janus>
    Gemma 3 <gemma3>
+   Llama 4 <llama-4>
    Llama 3.2 <llama-3_2>
    Llama 3.1 <llama-3_1>
    Llama 3 <llama-3>
diff --git a/docs/source/examples/models/llama-4.md b/docs/source/examples/models/llama-4.md
new file mode 120000
index 00000000000..fd8197c8f79
--- /dev/null
+++ b/docs/source/examples/models/llama-4.md
@@ -0,0 +1 @@
+../../../../llm/llama-4/README.md
\ No newline at end of file
diff --git a/llm/llama-4/README.md b/llm/llama-4/README.md
new file mode 100644
index 00000000000..63d0904e900
--- /dev/null
+++ b/llm/llama-4/README.md
@@ -0,0 +1,156 @@
+
+<!-- $REMOVE -->
+# Run Llama 4 on Kubernetes or Any Cloud
+<!-- $END_REMOVE -->
+<!-- $UNCOMMENT# Llama 4 -->
+
+
+[Llama 4](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) family was released by Meta on Apr 5, 2025.
+
+![](https://i.imgur.com/kjqLX87.png)
+
+
+## Prerequisites
+
+- Go to the [HuggingFace model page](https://huggingface.co/meta-llama/) and request access to the model [meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8](https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8).
+- Check that you have installed SkyPilot ([docs](https://docs.skypilot.co/en/latest/getting-started/installation.html)).
+- Check that `sky check` shows clouds or Kubernetes are enabled.
+
+## Run Llama 4
+
+```bash
+sky launch llama4.yaml -c llama4 --env HF_TOKEN
+```
+
+https://github.com/user-attachments/assets/48cdc44a-31a5-45f0-93be-7a8b6c6a0ded
+
+
+The `llama4.yaml` file is as follows:
+```yaml
+envs:
+  MODEL_NAME: meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8
+  HF_TOKEN: # TODO: Fill with your own huggingface token, or use --env to pass.
+
+resources:
+  accelerators: { H100:8, H200:8, B100:8, B200:8, GB200:8 }
+  cpus: 32+
+  disk_size: 512  # Ensure model checkpoints can fit.
+  disk_tier: best
+  ports: 8081  # Expose to internet traffic.
+
+setup: |
+  uv pip install vllm==0.8.3
+
+run: |
+  echo 'Starting vllm api server...'
+
+  vllm serve $MODEL_NAME \
+    --port 8081 \
+    --tensor-parallel-size $SKYPILOT_NUM_GPUS_PER_NODE \
+    --max-model-len 430000
+
+```
+
+You can use other models by setting different `MODEL_NAME`.
+```bash
+sky launch llama4.yaml -c llama4 --env HF_TOKEN --env MODEL_NAME=meta-llama/Llama-4-Scout-17B-16E-Instruct
+```
+
+
+🎉 **Congratulations!** 🎉 You have now launched the Llama 4 Maverick Instruct LLM on your infra.
+
+### Chat with Llama 4 with OpenAI API
+
+To curl `/v1/chat/completions`:
+```console
+ENDPOINT=$(sky status --endpoint 8081 llama4)
+
+curl http://$ENDPOINT/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
+    "messages": [
+      {
+        "role": "system",
+        "content": "You are a helpful assistant."
+      },
+      {
+        "role": "user",
+        "content": "Who are you?"
+      }
+    ]
+  }' | jq .
+```
+
+To stop the instance:
+```console
+sky stop llama4
+```
+
+To shut down all resources:
+```console
+sky down llama4
+```
+
+## Serving Llama-4: scaling up with SkyServe
+
+
+With no change to the YAML, launch a fully managed service on your infra:
+```console
+HF_TOKEN=xxx sky serve up llama4.yaml -n llama4 --env HF_TOKEN
+```
+
+Wait until the service is ready:
+```console
+watch -n10 sky serve status llama4
+```
+
+<details>
+<summary>Example outputs:</summary>
+
+```console
+Services
+NAME   VERSION  UPTIME  STATUS  REPLICAS  ENDPOINT
+llama4 1        35s     READY   2/2       xx.yy.zz.100:30001
+
+Service Replicas
+SERVICE_NAME  ID  VERSION  IP            LAUNCHED     RESOURCES                  STATUS  REGION
+llama4        1   1        xx.yy.zz.121  18 mins ago  1x GCP([Spot]{'H100': 8})  READY   us-east4
+llama4        2   1        xx.yy.zz.245  18 mins ago  1x GCP([Spot]{'H100': 8})  READY   us-east4
+```
+</details>
+
+
+Get a single endpoint that load-balances across replicas:
+```console
+ENDPOINT=$(sky serve status --endpoint llama4)
+```
+
+> **Tip:** SkyServe fully manages the lifecycle of your replicas. For example, if a spot replica is preempted, the controller will automatically replace it. This significantly reduces the operational burden while saving costs.
+
+To curl the endpoint:
+```console
+curl http://$ENDPOINT/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
+    "messages": [
+      {
+        "role": "system",
+        "content": "You are a helpful assistant."
+      },
+      {
+        "role": "user",
+        "content": "Who are you?"
+      }
+    ]
+  }' | jq .
+```
+
+To shut down all resources:
+```console
+sky serve down llama4
+```
+
+See more details in [SkyServe docs](https://docs.skypilot.co/en/latest/serving/sky-serve.html).
+
diff --git a/llm/llama-4/llama4.yaml b/llm/llama-4/llama4.yaml
new file mode 100644
index 00000000000..bdce6080b3a
--- /dev/null
+++ b/llm/llama-4/llama4.yaml
@@ -0,0 +1,34 @@
+envs:
+  MODEL_NAME: meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8
+  # MODEL_NAME: meta-llama/Llama-3.2-3B-Vision
+  HF_TOKEN: # TODO: Fill with your own huggingface token, or use --env to pass.
+
+resources:
+  accelerators: { H100:8, H200:8, B100:8, B200:8, GB200:8 }
+  cpus: 32+
+  disk_size: 512  # Ensure model checkpoints can fit.
+  disk_tier: best
+  ports: 8081
+
+setup: |
+  uv pip install "vllm>=0.8.3"
+
+run: |
+  echo 'Starting vllm api server...'
+
+  vllm serve $MODEL_NAME \
+    --port 8081 \
+    --tensor-parallel-size $SKYPILOT_NUM_GPUS_PER_NODE \
+    --max-model-len 430000
+
+service:
+  replicas: 2
+  # An actual request for readiness probe.
+  readiness_probe:
+    path: /v1/chat/completions
+    post_data:
+      model: $MODEL_NAME
+      messages:
+        - role: user
+          content: Hello! What is your name?
+      max_tokens: 1