CVE-2026-105758: vLLM Memory Exhaustion Vulnerability in Qwen Video Samplers
Summary
CVE-2026-105758: An unauthenticated remote attacker can exhaust the memory of the vLLM API-server process by raising the request-level media_io_kwargs.video.max_frames and fps knobs on any deployment serving a Qwen2-VL or Qwen3-VL model. Affects vllm versions >= 0.24.0, < 0.30.0. Fixed in 0.30.0. How to fix ↓
Check if this vulnerability affects your application
Upload a manifest or SBOM and we'll show whether you're affected, plus every other risky package in the same scan.
No signup required · Results in about 60 seconds
Habilume
Habit tracker with a companion that grows with you
Complete habits, earn Lume Points, and watch your Lume grow as you stay consistent.
Background
vLLM is an open-source library designed for efficient video processing and model serving, particularly for large language models and video samplers like Qwen2-VL and Qwen3-VL. These models are widely used for various applications, including video analysis and generation. However, a critical vulnerability has been identified that allows unauthenticated remote attackers to exhaust the memory of the vLLM API-server process, leading to potential denial of service.
Vulnerability Detail
The vulnerability arises from the lack of proper limits on request-controlled parameters, specifically media_io_kwargs.video.max_frames and fps, in the Qwen video samplers. An attacker can exploit this by sending requests with excessively high values for these parameters, causing the server's memory usage to spike dramatically. This results in memory exhaustion, affecting all tenants on the instance. The vulnerability is particularly severe as it allows unauthenticated access to the API, making it easy for attackers to exploit it without any authentication barriers.
How To Fix
Latest published version: 0.32.0 as of 22 Oct 2026
pip install --upgrade "vllm==0.32.0"Upgrading to it may include breaking changes; the version below is the minimum fixed release for this vulnerability.
If you need the minimum change
0.30.0
This is the first release that contains the fix.
pip install --upgrade "vllm==0.30.0"Tip — Not sure which version your app runs? Vulert can match this fix to it. Scan free →
Workaround
As a temporary workaround, you can limit the values of max_frames and fps in your requests to prevent excessive memory usage. For example, ensure that max_frames does not exceed 768 and fps is kept at a reasonable level. This can be done by modifying your request payload as follows:
{
"media_io_kwargs": {
"video": {
"max_frames": 768,
"fps": 30
}
}
}
Conclusion
In summary, CVE-2026-105758 is a significant vulnerability in the vLLM package that allows for memory exhaustion through unauthenticated requests. It is crucial for users to update to the patched version 0.30.0 to mitigate this risk. Additionally, temporary workarounds can be implemented to limit the impact until the update is applied.
See every vulnerability recorded for vllm →
FAQ
The CVSS score of this vulnerability is 5.3, indicating a moderate severity level.
You can use Vulert's playground at Vulert Scanner to check if your application is affected by this vulnerability without any signup.
The potential impacts include denial of service due to memory exhaustion, affecting all users of the vLLM API on the affected instance.
Yes, you can mitigate the vulnerability by limiting the max_frames and fps parameters in your requests to prevent excessive memory usage.
The vulnerability affects vLLM versions >= 0.24.0 up to v0.29.1rc0. It is fixed in version 0.30.0.
References
Related vulnerabilities
Recent vulnerabilities recorded for vllm.
Habilume
Habit tracker with a companion that grows with you
Complete habits, earn Lume Points, and watch your Lume grow as you stay consistent.
