CVE-2026-126397 vLLM 0.10.1 before 0.31.0 Cohere2 Vision DoS via max_patches
HighCVSS 8.7
EPSS 0.46%
- Published
- Updated
- Exploit status
- No public exploit known
- CISA SSVC
- No known exploitation · automatable: yes · technical impact: partial
Description
vLLM from 0.10.1 before 0.31.0 forwards a request-controlled mm_processor_kwargs.max_patches value to the Hugging Face processor for Cohere2 Vision models without a server-side upper bound, causing quadratic CPU work during multimodal preprocessing. Remote clients can send small image chat-completion requests with large, varying max_patches values to bypass the LRU cache, exhaust preprocessing workers, and deny service to other clients.
Upgrade path
1 package: 1 fixed.
| Package | Affected | Fixed in | Status |
|---|---|---|---|
| pypi vllm | >= 0.10.1, < 0.31.0 | 0.31.0 | Fixed |
Affected products
| Product | Affected versions | Fixed in |
|---|---|---|
| pypi vllm | >= 0.10.1, < 0.31.0 | 0.31.0 |
References
- GitHub Security Advisory (GHSA-9g3c-x747-c6wr) vendor-advisory
- Patch Commit patch
- VulnCheck Advisory: vLLM 0.10.1 before 0.31.0 Cohere2 Vision DoS via max_patches third-party-advisory
- ENISA EUVD EUVD-2026-125168 third-party-advisory