Tencent released WeVisOne-VL 2B/4B weights, the pixel decoder and inference code.
Tencent releases WeVisOne-VL weights for unified visual perception
2026-10-23Verified announcementTRIAL
Tencent released WeVisOne-VL 2B and 4B model weights plus a required 9.6M-parameter pixel decoder for detection, referring localization, counting, point grounding, OCR, segmentation and alpha matting. The Hugging Face repositories are ungated and list Apache-2.0, while the GitHub detector does not expose a detected license. Published benchmarks are vendor-reported.
Availability: ungated Hugging Face weights and public inference code
What this means for you
For teams that currently chain separate vision models, one open-weight interface may simplify experimentation and deployment. The broad task surface also raises evaluation risk: license notices, pixel-decoder packaging and accuracy on local images require independent checks.
Evidence & related material
Technical details & scores
| Base | Qwen3-VL |
|---|---|
| Sizes | 2B and 4B plus required 9.6M pixel decoder |
| Tasks | open-vocabulary detection, referring localization, counting, point grounding, OCR, referring segmentation and alpha matting |
| Availability | ungated Hugging Face weights and public inference code |
| License Boundary | model cards list Apache-2.0; GitHub license detector reports NOASSERTION, so inspect included third-party notices |
| Benchmark Boundary | publisher-reported only |
| Time Basis | public GitHub release commit at 2026-10-23T08:14:47Z; Hugging Face repositories expose date-only creation evidence |
Technical impact 4
Engineering value 4
Adoption signal 2
Maturity 2
Verification cost 3
Risk 3