NVIDIA is preparing Nemotron Parse 3.5 📄 A vLLM PR adds nvidia/Nemotron-3.5-Parse, not yet public on Hugging Face. A redesign of their document parser: • causal visual prefix, no cross-attention • GQA instead of MHA • dynamic-resolution vision encoder • built-in MTP head
Leaked, then released ·
nvidia/Nemotron-3.5-Parse
by NVIDIA found in NVIDIA Nemotron Parse 3.5 vLLM PR
New model
Leaked first
- Status
- Released Oct 17, 2026, leaked 3 days earlier NVIDIA Nemotron official model weights
- Maker
- NVIDIA
- Found in
- NVIDIA Nemotron Parse 3.5 vLLM PR github.com
- Detected
- Posted to X
- 0 s after detection
- Leak lead
- 3 daysbefore the release
- Model ID
nvidia/Nemotron-3.5-Parse
Where to use it
first seen Oct 14, 2026| Where | Listed | After first sighting |
|---|---|---|
| Hugging Face weights | +3 days |
Timeline
1 sighting-
reference
Spotted in NVIDIA Nemotron Parse 3.5 vLLM PR
NVIDIA is preparing Nemotron Parse 3.5 📄
Post on X -
release
Released in NVIDIA Nemotron official model weights