# Furiosa-LLM 2026.5.0 Beta4, 2026.4.1, and 2026.3.2 are now available!

**URL:** <https://forums.furiosa.ai/t/furiosa-llm-2026-5-0-beta4-2026-4-1-and-2026-3-2-are-now-available/502>\
**Category:** Furiosa LLM\
**Tags:** sdk, rngd, release\
**Created:** [November 4, 2026, 3:33am UTC](https://forums.furiosa.ai/t/furiosa-llm-2026-5-0-beta4-2026-4-1-and-2026-3-2-are-now-available/502 "2026-11-04T03:33:14Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![hyunsik](https://yyz1.discourse-cdn.com/flex031/user_avatar/forums.furiosa.ai/hyunsik/32/6_2.png) [@hyunsik](https://forums.furiosa.ai/u/hyunsik)\
**Post date:** [November 4, 2026, 3:33am UTC](https://forums.furiosa.ai/t/furiosa-llm-2026-5-0-beta4-2026-4-1-and-2026-3-2-are-now-available/502/1 "2026-11-04T03:33:14Z")

</div>

Hi everyone!

2026.5.0 Beta4, 2026.4.1, and 2026.3.2 are now available on the public Python package repository.

**2026.5.0 Beta4** — improvements since Beta3

- Implemented repetition\_detection for the OpenAI-compatible Completions and Chat Completions APIs.
- Added return\_prompt\_text for Chat Completions and support for interleaved reasoning in chat templates.
- JSON-schema structured outputs now disable extra whitespace by default.
- Improved KV-cache allocation for long-context serving, with clearer startup errors when the requested context length exceeds available capacity.
- Improved native multimodal request handling and fixed sampled-token and vision-buffer allocation issues.

**2026.4.1 and 2026.3.2** — security hotfix  
• eMudhra-related certificate trust updates and Python dependency security fixes

* * *

### Installation

Choose the version for your release line:

```bash
# 2026.5.0 beta4
uv pip install --prerelease=allow --torch-backend auto furiosa-llm==2026.5.0b4

# Latest stable
uv pip install --torch-backend auto furiosa-llm==2026.4.1

# 2026.3 release
uv pip install --torch-backend auto furiosa-llm==2026.3.2

```
