GigaChat 3 Ultra vs. BaseRT

GigaChat 3 Ultra is a breakthrough open-source LLM, offering 702 billion parameters built on an advanced MoE architecture that keeps computation efficient while delivering frontier-level performance. Its design activates only 36 billion parameters per step, combining high intelligence with practical deployment speeds, even for research and enterprise workloads. The model is trained entirely from scratch on a 14-trillion-token dataset spanning ten+ languages, expansive natural corpora, technical literature, competitive programming problems, academic datasets, and more than 5.5 trillion synthetic tokens engineered to enhance reasoning depth. This approach enables the model to achieve exceptional Russian-language capabilities, strong multilingual performance, and competitive global benchmark scores across math (GSM8K, MATH-500), programming (HumanEval+), and domain-specific evaluations. GigaChat 3 Ultra is optimized for compatibility with modern open-source tooling, enabling fine-tuning, inference, and integration using standard frameworks without complex custom builds. Advanced engineering techniques—including MTP, MLA, expert balancing, and large-scale distributed training—ensure stable learning at enormous scale while preserving fast inference. Beyond raw intelligence, the model includes upgraded alignment, improved conversational behavior, and a refined chat template using TypeScript-based function definitions for cleaner, more efficient interactions. It also features a built-in code interpreter, enhanced search subsystem with query reformulation, long-term user memory capabilities, and improved Russian-language stylistic accuracy down to punctuation and orthography. With leading performance on Russian benchmarks and strong showings across international tests, GigaChat 3 Ultra stands among the top five largest and most advanced open-source LLMs in the world. It represents a major engineering milestone for the open community.

BaseRT provides a powerful inference runtime for large language models, specifically tailored for Apple Silicon, enabling developers to effortlessly access models from Hugging Face, engage in local dialogue, or use an OpenAI-compatible API all through a single command-line interface. Boosted by expertly designed Metal kernels, BaseRT is engineered to excel in prefill and decoding efficiency on M-series Macs, with benchmark tests demonstrating performance that is up to 6.4 times faster in prefill tasks than llama.cpp, 3.9 times quicker than MLX, and achieving a decoding speed that surpasses competitors by 1.33 times. The BaseRT CLI is equipped to handle various tasks including model downloading, conversion, interactive chatting, serving functionalities, completion generation, benchmarking, inspection, and bundle signing. Its comprehensive server capabilities include chat interactions, text completions, embeddings, transcription services, tool calls, continuous batching, paged key-value caching, and prefix caching, while supporting models that process text, vision, and audio data. BaseRT utilizes a unique .base model format that features Q2–Q8 affine quantization, optional AWQ calibration, and signed bundles, and it can convert GGUF, Hugging Face, and MLX checkpoints seamlessly. In addition to these features, this groundbreaking runtime is specifically designed to harness the full potential of Apple Silicon, establishing itself as an indispensable resource for developers working in the AI domain. With its impressive efficiency and broad functionality, BaseRT stands out as a key innovation for the future of AI development on Apple platforms.