# DeepSeek publishes open weights for V4 vision model

_Published Tuesday, September 1, 2026 at 10:19 AM EDT · AI · Latest · Tier 1 — Major_

![aerial photograph taken January 14 2026 shows — Primary](https://d.techtimes.com/en/full/474267/aerial-photograph-taken-january-14-2026-shows.jpg)

DeepSeek published open weights for V4-Flash-Vision-Exp on Aug. 31, adding image capability to its V4 model family, according to the report. The 305-billion-parameter sparse mixture-of-experts model is downloadable under an MIT license, allowing self-hosting, fine-tuning and deployment without a license fee. The report says the model supports image inputs and retains a one-million-token context window, but its vendor benchmark results have not been independently reproduced.

## Sources

- [Tech Times](https://www.techtimes.com/articles/326132/20260901/deepseek-open-sources-first-v4-vision-model-benchmark-claims-need-independent-proof.htm)

---
Canonical: https://techandbusiness.org/newswire/CawTeWvDlwwxe5hK_m_6G9
Published: 2026-09-01T14:19:29.436Z
Story chronology: 2026-08-31T00:00:00.000Z
Retrieved: 2026-10-17T03:11:29.414Z
Publisher: Tech & Business (techandbusiness.org)