Models & Labs
Moonshot releases largest open AI model K3
The Rundown AIhigh confidence
Why it matters
- →Moonshot's release of K3 could democratize access to advanced AI models.
- →It challenges the current trend of restricted access to frontier AI models.
- →This move may influence future regulatory discussions on AI model accessibility.
Moonshot has released the weights for its Kimi K3 model, making it the largest open AI model available. The 2.8 trillion parameter model requires significant GPU resources to run, but Moonshot has also opened key components of its stack for broader use. This release could change the landscape of AI accessibility, as it challenges the current trend of restricted access to advanced models. The move comes amid U.S. concerns about the competitive edge of Chinese open models.
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 from The Rundown AI
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.
