Open Source
llama.cpp adds IBM ZDNN backend support
llama.cpp Releaseshigh confidence
Why it matters
- →Enables efficient LLM inference on IBM Z mainframes via ZDNN.
- →Expands llama.cpp's hardware support to s390x architecture.
- →Signals growing enterprise interest in diverse AI infrastructure.
The llama.cpp project has merged a pull request adding continuous integration builds for the IBM ZDNN backend, targeting s390x architecture. Signed off by IBM engineer Aaron Teo, the update includes compiler fixes and CI configuration adjustments to support Ubuntu-based builds on mainframe hardware. Although automated testing is currently disabled, the inclusion of this backend in the official release pipeline marks a significant step toward broader hardware compatibility for local LLM inference beyond standard x86 and ARM platforms.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
How we got here
- 1llama.cpp adds Qwen3.8-Flash-Next support
llama.cpp Releases · August 28, 2026 · Related
- 2llama.cpp b10970 Release Expands Platform Support
llama.cpp Releases · September 15, 2026 · Related
More from llama.cpp Releases
Coding Toolscoding
llama.cpp b12336 fixes HunyuanVL crash
This release quietly patches a critical memory overflow in the mtmd module that caused crashes when loading HunyuanVL models. The issue stemmed from F64 position embeddings overflowing an F32 buffer, a subtle bug that broke multimodal inference for users relying on this specific architecture. By enforcing strict F32 requirements, llama.cpp stabilizes support for Tencent's vision-language model without adding new features or changing the core API. It’s a necessary maintenance update that keeps local multimodal workflows from breaking unexpectedly.
llama.cpp Releases
Models & Labsother
llama.cpp b12340 adds CUDA 13 and ROCm 10.0
This release quietly closes the hardware gap for local inference by adding default support for CUDA 13 and ROCm 10.0 across Linux and Windows. Users with newer NVIDIA GPUs or AMD accelerators no longer need to compile from source to get performance, while Apple Silicon builds now ship KleidiAI disabled by default, shifting the focus back to standard ARM optimizations. The inclusion of Snapdragon support for Linux and Android marks a significant step toward on-device AI beyond the x86/ARM desktop ecosystem. It is less about new features and more about ensuring llama.cpp remains the universal runtime for every available accelerator.
llama.cpp Releases
Models & Labscoding
llama.cpp b12342 adds few-row MMQ for speculative decoding
Speculative decoding just got a significant speed boost on NVIDIA GPUs. This release introduces a specialized kernel for 'few-row' matrix multiplications, targeting the small batch sizes typical of verification passes in speculative decoding workflows. By optimizing these specific operations for quantized formats like Q8_0 and IQ4 variants, llama.cpp reduces latency where it matters most during token generation. The change is CUDA-only for now but leverages hardware capabilities from Ampere onwards, making local inference faster without requiring new model architectures or complex configuration changes.
llama.cpp Releases
More in Open Source
© Hugging Face BlogOpen Sourceagents
Tether AI launches BotPsy, a sub-billion agentic model
Tether AI Research has released BotPsy, an 0.8B-parameter model built on Qwen3.5 that specializes in tool-use and agentic workflows for on-device deployment. It achieves state-of-the-art performance in its size class, outperforming larger models like Qwen3.5-2B-Think on benchmarks such as BFCL v4 and ACEBench while maintaining strict privacy by keeping data local. The model is designed to handle routine tasks like calendar management and web search without relying on expensive cloud APIs, offering a viable path for private, self-hosted personal assistants. With open weights under Apache 2.0 and GGUF support, it provides immediate utility for developers building lightweight, autonomous agents.
Hugging Face Blog
© Hugging Face BlogOpen Sourceother
NVIDIA releases 9M synthetic Japanese finance personas
NVIDIA has released a massive synthetic dataset of 9 million Japanese financial personas, addressing the privacy and data scarcity bottlenecks in localizing AI for Japan. By grounding these profiles in official government statistics rather than real customer records, it allows developers to train and red-team agents without violating strict privacy regulations. The inclusion of specific attributes like household assets and financial literacy provides a level of demographic granularity that generic Western-centric datasets simply cannot match. This effectively lowers the barrier for building culturally accurate, compliant AI systems in one of the world's largest economies.
Hugging Face Blog
© Lev SelectorOpen Sourceimage
FLUX 3 Image Klein Open Weights Released
Black Forest Labs released FLUX 3 Image Klein, an open-weight image generation model offering high-quality results for developers.
Lev Selector