# Deepseek V4 released

**URL:** <https://forums.developer.nvidia.com/t/deepseek-v4-released/367696>\
**Category:** DGX Spark / GB10\
**Tags:** deepseek\
**Created:** [April 24, 2026, 3:21am UTC](https://forums.developer.nvidia.com/t/deepseek-v4-released/367696 "2026-04-24T03:21:32Z")\
**Posts on this page:** 1\
**Showing post:** 89

<div class="post-metadata">

**Author:** ![arthurdroz](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/arthurdroz/32/506945_2.png) [@arthurdroz](https://forums.developer.nvidia.com/u/arthurdroz)\
**Post date:** [April 29, 2026, 12:43pm UTC](https://forums.developer.nvidia.com/t/deepseek-v4-released/367696/89 "2026-04-29T12:43:31Z")

</div>

I created an experimental eugr recipe PR here, seeing some issues with tool calling but getting around 22 t/s [Add DeepSeek V4 Flash recipe by arthur-drozdov · Pull Request #219 · eugr/spark-vllm-docker · GitHub](https://github.com/eugr/spark-vllm-docker/pull/219)

---

_[View the full topic](https://forums.developer.nvidia.com/t/deepseek-v4-released/367696)._