# NVIDIA releases TensorRT‑LLM 1.4.0 release candidate

> The new release adds support for several models, a sparse KV cache, and a Mooncake store pool, while fixing bugs around multi‑GPU streaming and quantized heads.

Oossa · 2026-10-14 · https://oossa.com/en/nvidia-releases-tensorrt-llm-1-4-0-release-candidate

NVIDIA announced TensorRT‑LLM version 1.4.0rc0 on Oct 14, 2026. The update brings new model support like GLM‑5.3‑Flash and Qwen‑Image‑Layered, and introduces a sparse key‑value cache with tier‑aware locking. It also adds a Mooncake store pool and a V2 scheduler preemption feature. Several bugs were fixed, including hangs in multi‑GPU streaming on H100 GPUs and crashes with DeepSeek models on Blackwell GPUs.

## The facts

- Release date: 2026‑10‑14 (Wed Oct 14 2026) – from the GitHub tag page
- New features: sparse KV cache, Mooncake store pool, V2 scheduler preemption

## Why it matters

Developers can run larger language models on NVIDIA GPUs with lower memory use and fewer crashes.

## Sources & references

1. [NVIDIA/TensorRT-LLM v1.4.0rc0](https://github.com/NVIDIA/TensorRT-LLM/releases/tag/v1.4.0rc0) – TensorRT-LLM, 2026-10-14

Last updated: 2026-10-14