Models & Labs
Tencent Unveils Hy3 Model as GLM Competitor
Sam Witteveenhigh confidence
Why it matters
- →Hy3 offers a more efficient alternative to GLM 5.2, being half its size.
- →It expands Tencent's presence in the AI model landscape with a focus on non-code tasks.
- →Developers gain a new tool for diverse applications, potentially enhancing AI utilization.
Tencent has launched Hy3, a new AI model designed to compete with GLM 5.2. Hy3 is a mixture of experts model that is notably half the size of GLM 5.2, yet it maintains strong performance across a range of tasks, excluding those related to coding. This release underscores Tencent's commitment to advancing AI technology by providing a more efficient model for various applications. The introduction of Hy3 offers developers and researchers a new option for leveraging AI in non-code tasks.
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.
More in Models & Labs
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.
