GLM-Z1-Rumination-32B-0414 is a reasoning-capable model with 32 billion parameters, specifically trained to solve complex research and analytical tasks, with the ability to use external search. It excels at engaging in prolonged deliberation, allowing it to effectively handle multi-step assignments.
GLM-4-32B-0414 is a powerful model with 32 billion parameters, trained on 15 TB of high-quality data. In terms of performance, it is comparable to leading models such as GPT-4o and DeepSeek-V3-0324, particularly in programming tasks, while remaining lightweight for easy local deployment.
Llama 4 Scout is a model with native multimodality and a context window of up to 10 million tokens, while running on a single GPU. It is ideal for analyzing large text arrays and quickly extracting information from images.
Llama 4 Maverick - supports context windowing up to 1 million tokens, native multimodality and demonstrates high speed and efficiency due to the combination of 128 experts and 400 billion parameters in the architecture. The model is well suited for programming and technical documentation tasks.
A specialized Russian-language model with 8B parameters and a 32k token context, trained entirely from scratch on Russian and English data. Thanks to optimized tokenization and innovative training techniques, the model outperforms similar-sized solutions like Llama and Qwen, especially in tasks related to Russian culture and language.
DeepSeek-V3 0324 is an enhanced version of DeepSeek's powerful and popular MoE model with 685 billion parameters. Demonstrates exceptional quality, deep answer mining, and outstanding erudition in tasks ranging from analyzing complex legal documents to generating executable program code from scratch.
Gemma 3 27B — flagship multimodal model from Google DeepMind with 27 billion parameters and maximum performance. It is easy to fine-tune and ideal for a wide range of complex research tasks and high-end enterprise solutions.
Gemma 3 1B — an ultra-compact model with just 1 billion parameters, yet retaining impressive capabilities. It supports a context window of 32K tokens and is ideal for resource-constrained devices and tasks where response speed is critical.
Gemma 3 12B is a high-performance multimodal model with 12 billion parameters, a context window of 128K tokens, and multilingual understanding, designed for a wide range of straightforward tasks. It excels at processing long documents, images, and technical content.
Gemma 3 4B is a compact model that is also multimodal, featuring a context window of 128K tokens and built-in support for more than 35 languages, including Russian. It's an excellent solution for embedded systems and applications processing text and images with limited computational resources.
QwQ is a model with 32.5 billion parameters and a context length of 40K tokens, specifically designed for deep reasoning and logical analysis. Its unique ability to perform transparent and structured thinking sets it apart from competitors, delivering high-quality and well-thought-out responses.
This is a Text-to-Video model with 1.3 billion parameters, developed for generating video from text prompts. The model is optimized for consumer-grade GPUs: it requires 8.19 GB of VRAM, and generating a 5-second video in 480p resolution takes ~4 minutes on an RTX 4090 GPU without optimization.
Phi-4-multimodal is an efficient solution for multimodal tasks with edge deployment support, combining a compact size (5.6B parameters) with the capabilities of large language models. The model is ideal for developing applications with synchronous processing of speech, images, and text on resource-constrained devices.
Chroma is an 8.9 billion-parameter model based on the FLUX.1-schnell architecture.
Qwen2.5-VL-7B is a powerful multimodal model with 7 billion parameters, delivering an optimal balance between high performance and efficiency. Designed for complex document analysis, video stream processing, and agent-based interaction tasks.
Qwen2.5-VL-3B - is a compact, 3-billion-parameter multimodal model designed for edge deployment, yet it delivers outstanding capabilities in image/video comprehension and agent-based task execution.
Qwen2.5-7B-1M is a compact yet powerful model with 7.6 billion parameters. Thanks to sparse attention technologies, it can process up to one million context tokens at excellent speeds. The model is an ideal solution for organizations requiring high-performance analysis of long documents while optimizing resource usage.
DeepSeek-R1-Distill-32B — a model built based on distilling a large MoE reasoning expert-level model, setting new records among open-source dense models. It is suitable for scientific, corporate, and educational platforms with high demands on logic and analysis.
DeepSeek-R1 is a unique reasoning model with 671 billion parameters, trained based on reinforcement learning (RL), supporting long chains of thought (CoT), and specializing in multi-step reasoning and logical analysis. It is indispensable for tasks requiring well-founded conclusions and transparent reasoning processes.
DeepSeek-R1-Distill-1.5B — a compact model that, thanks to distillation, possesses strong reasoning capabilities. It is ideal for fast text analysis in mobile and edge applications.