# llama.cpp update adds reliable realtime audio streaming

> The Oct 23 2026 release improves the realtime API so audio streams finish only after all committed data is decoded.

Oossa · 2026-10-22 · https://oossa.com/en/llama-cpp-update-adds-reliable-realtime-audio-streaming

The ggml‑org/llama.cpp project released version b11828 on Oct 23 2026. The main change is to the realtime API: a commit is now marked complete only after the audio buffer fully drains, preventing partial playback. The update also lists new builds for macOS, Linux, Windows, Android and more, including CUDA 13 and Vulkan support.

## The facts

- Release tag b11828 published on 2026-10-23
- Realtime API now waits for full audio drain before completing a commit

## Why it matters

Developers can build voice‑enabled apps with llama.cpp that play back audio without cut‑offs.

## Sources & references

1. [ggml-org/llama.cpp b11828](https://github.com/ggml-org/llama.cpp/releases/tag/b11828) – llama.cpp, 2026-10-22

Last updated: 2026-10-22