CVE-2026-126396
MiniMax M3 video resizing could exhaust inference workers
vLLM · MiniMax M3 video processor
First documented disclosure · Review checkpoint
Summary
The video processor accepted a client-defined minimum image size without enforcing a total pixel budget by default. Resizing decoded frames could consume enough CPU and memory to disrupt the worker or service.
Severity and exploitation evidence
Medium · 6.9/10 · CVSS 4.0. Severity source.
Exploitation not established: The reviewed sources describe the weakness and correction. They identify no confirmed malicious use.
Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.
How the weakness works
The video processor accepted a client-defined minimum image size without enforcing a total pixel budget by default.
Conditions required
The service runs the affected vision model and permits client-supplied preprocessing settings.
Potential impact
Resizing decoded frames could consume enough CPU and memory to disrupt the worker or service.
Affected versions and configurations
0.24.0 through releases before 0.31.0.
Fix and mitigation status
0.31.0 for this issue.
Practical next steps
Use a supported vLLM release containing the 0.31.0 fix. Keep client-controlled multimodal settings disabled for untrusted users.
Operating lesson
Enforce a total budget across all frames as well as a limit on any single frame.
Evidence and limits
A September mitigation covered a class of multimodal settings. The October project notice describes this processor-specific exposure. The issuing CNA published the CVE record on 2026-11-13.
Sources
- vLLM project security advisory
Primary source ·
- Issuing CNA record and severity assessment
Primary source ·