New model: GigaChat3.5-432B-A28B (with day-0 GGUF support!)
Sberbank released GigaChat3.5-432B-A28B, a new large language model with 432 billion parameters, and provided day-zero GGUF support for efficient local inference.
- GigaChat3.5-432B-A28B is a 432B parameter LLM from Sberbank, one of the largest open models released to date.
- Day-zero GGUF support allows efficient local inference, reducing barriers to adoption for developers.
- The GGUF version is available via a pull request in the llama.cpp repository, not yet in the main branch.
- This release underscores Sberbank's expanding role in the open-source AI community.
Sberbank has introduced GigaChat3.5-432B-A28B, a new large language model with 432 billion parameters, positioning it among the largest open models available. The model is designed for high-performance applications and includes a base version for fine-tuning. Notably, the team has provided day-zero GGUF support, enabling efficient local inference without requiring complex setups. This is significant because GGUF is a widely adopted format for running LLMs on consumer hardware, making the model more accessible to developers and researchers. The GGUF version is not yet in the main branch but can be built from a pull request in the llama.cpp repository, indicating an active community-driven effort to integrate the model quickly. The release reflects Sberbank's growing influence in the open-source AI space, following its previous model launches and contributions to the ecosystem.
Source: New model: GigaChat3.5-432B-A28B (with day-0 GGUF support!). Read the full piece at the source.
Enables local deployment of a 432B parameter model with GGUF, reducing infrastructure costs and improving accessibility.
Demonstrates the growing competition among organizations to release large, open models with practical deployment options.
- GGUF
- A file format for quantized large language models, enabling efficient local inference on consumer hardware.
New open model from Tencent Hy: Hy3 (295B total 21B active - apache 2.0)
Doing the actual math on a $20k local AI rig breakeven
