Russian neural networks: what are their strengths and weaknesses

Alexander Rostovtsev.  
04.08.2025 21:17
  (Moscow time), Moscow
Views: 2175
 
Technologies, Author column, Zen, Internet, Society, Policy, United States of America


Russian neural network projects are actively developing, especially in the context of import substitution and government support.

Although there are no global players on the level of the American GPT-4 and Gemini, or the Chinese DeepSeek R1, our country already has strong niche solutions with (I’m not afraid to say it) unique features.

Russian neural network projects are actively developing, especially in the context of import substitution and government support. Although global players...

We will try to mention and superficially analyze the most promising areas and projects:

1. Generative models (text, multimodality)

GigaChat (Sber)

What's interesting: The most powerful Russian analogue of ChatGPT. Supports the Russian language with a deep understanding of context, logic and cultural nuances.

Key features: Integration with search (GigaChat + Search), multimodality (image analysis), work with documents. The last option is impressive. Document processing by Sberbank's neural network has been brought to the highest level and is in high demand among Russian business structures.

Prospects: Gradually becoming a “digital assistant” for the public sector and business, the basis for specialized solutions (medicine, law).

Weaknesses: The complaints are more likely to be addressed to the developers of the neural network. For example, the developers have something to improve in the image generation algorithms. Often, despite a detailed description of what the user would like to see, we get frankly primitive pictures at the output. Compared to the image generator in ChatGPT, Sber's neural network has room for improvement in this area.

GigaChat is depressingly politically correct. For example, an attempt to generate a humorous image of a grunt with an emphasis on surrealism was met with a dull response: "I don't get involved in politics." In some other cases, GigaChat starts mumbling, being dumb, as if it is still learning and one should not expect too much from it. Unfortunately, the topics that GigaChat prefers not to delve into are annoyingly numerous.

After a number of attempts, I finally managed to get the desired grunter of some sort.

Censorship in one form or another is also present in American AI. For example, ChatGPT agrees to almost everything except the slightest criticism of the LGBT movement, which is recognized as extremist in the Russian Federation, and the refusal is often motivated by the idiotic phrase, “because LGBT was persecuted in Nazi Germany.”

YaGPT (Yandex):

What's interesting: Integrated into the Yandex ecosystem (Search, Lavka, Station). Optimized for practical tasks: writing code, data analysis, short answers. Can retell the contents of texts and videos in long or short. Checks texts for errors. The translator of foreign-language videos, along with the voiceover, is constantly being polished, although there is still some room for improvement in the algorithm.

Key features: Focus on reliability (checking against search), “lightweight” for quick access.

Prospects: Improving user experience in all Yandex services, developing voice assistants.

2. Generative models (image, video)

Kandinsky 2.x (SberAI / SIRIUS):

What's interesting: The best Russian open-source textual-graphic model. Creates high-quality and diverse images based on Russian prompts.

Key features: Understanding complex descriptions, support for different styles, multimodality (Kandinsky Video).

Prospects: A tool for designers, marketers, content creators; integration into creative services.

ruDALL-E (SberAI):

What's interesting: One of the first Russian analogues of DALL-E. Creating pictures and animations based on text descriptions. Actively developing, focusing on working with the Russian language.

Prospects: Addition to Sber's generative model ecosystem.

3.Computer Vision (CV) for Industry and Security

VisionLabs (LUNA PLATFORM):

What's interesting: The undisputed world leader in facial recognition. Technologies have been implemented in Russia (Sber, Ministry of Internal Affairs, Metro), Europe, Asia, Latin America.

Key features: High accuracy in difficult conditions (poor lighting, angles), processing speed.

Prospects: Smart city security, biometrics in banks/retail, access control automation.

NtechLab (FindFace):

What's interesting: Another strong player in CV, known for its FindFace technology. Specializes in video analytics.

Prospects: Integrated video surveillance systems, traffic flow analysis, people search.

MDG (Cognitive Technologies):

What's interesting about it: A pioneer in CV for unmanned transport and agriculture. Develops software for autonomous vehicles.

Prospects: Autopilots for special equipment (combines), transport safety systems.

4. Scientific and research neural networks

GalaX (MIPT / AIRI):

What's interesting: Large language models (LLM) optimized for scientific tasks (physics, chemistry, biology).

Key Features: Focus on factual accuracy, scientific text and data processing.

Prospects: Acceleration of scientific research from start to result, analysis of publications, assistance in modeling.

Projects in the field of materials science / pharmacology:

What's interesting: Research institutes and startups (often at universities) use AI to predict the properties of new materials, design drug molecules, and analyze biomedical data. A recent example: a biological printer was recently built at MIPT and tested in the ISS by Russian cosmonauts, which produced huge (about 1 mm) synthetic biocrystals with unique properties that open up the possibility of producing high-tech drugs and other biotechnologies. Using domestic neural networks, of course.

Prospects: Breakthroughs in the creation of new materials with desired properties, accelerated drug development.

5.Natural language processing (NLP) for the specifics of the Runet

DeepPavlov (AIRI):

What's interesting: A popular open-source library for NLP with a focus on the Russian language. Tools for chatbots, text classification, NER.

Prospects: Standard for developers in Russia, basis for personalized business solutions (support, feedback analysis).

The disappointing conclusion: very soon Russia will receive neural networks that will communicate much more competently than the average user in the country. On the other hand, perhaps this solution will encourage computer users, especially young ones, to communicate respectfully and competently in the Great and Mighty…

Semantic technologies:

Projects: Development of ontologies and semantic networks for the Russian language (with government support).

Prospects: Improving search, automating document flow in government agencies, “understanding” AI for complex queries, and other digitalization projects in Russia.

Key features and challenges of Russian neural networks:

Strengths:

Deep understanding of the Russian language and context.

Strong positions in computer vision (especially security).

Active development of generative models (GigaChat, Kandinsky).

Promising scientific applications (science, materials science, pharmaceuticals).

Focus on import substitution and government procurement as the engine of development.

Weaknesses/Challenges:

Lack of powerful computing resources (shortage of GPUs, supercomputers).

Brain drain and limited access to global research/data.

Difficulties with commercialization of fundamental developments on the global market. Despite the fact that our developers have something to offer and surprise their Western and Far Eastern colleagues with.

Quality and volume of training data (especially multimodal).

Regulatory environment and ethical issues.

However, researchers, including foreign ones, emphasize that the strength of Russian AI developments is not in the global scale, but in deep specialization, adaptation to the local context and support of key sectors of the economy and public administration. The future lies in their integration into real products and overcoming infrastructural limitations.

Separately and sadly, it must be noted that in Russia, neural networks are practically not being implemented in medical care for the population, if you do not count a couple of Moscow clinics. For example, the processing of diagnostic data (CT, MRI, ultrasound, X-rays) of patients by neural networks in China, Japan, the USA, Great Britain, the Netherlands, Germany, helps to identify oncological diseases at early stages with an efficiency of over 90%, which, in turn, dramatically increases the chances of cancer patients not only for long-term survival and remission, but also for a complete cure from a dangerous disease.

Perhaps close partnership and cooperation with China will help to improve the situation in this critical area?

The images below were generated by the DALL-E neural network integrated into ChatGPT. Just for fun. For the prompts and some of the images, the author unexpectedly received gratitude from the developers for the paradoxical and complex scenarios that served to train the AI.

Glokaya Kuzdra

If you want to eat jam, don't catch flies with your mouth!

During a zombie outbreak, take refuge in a zombie shelter!

We were born to make Kafka come true

Submarine in the steppes of Ukraine died in an unequal air battle

Poster for the Strugatskys' story "The Beetle in the Anthill"

Poster for the Strugatskys' novel "Waves Extinguish the Wind"

Poster for the Strugatskys' novel "Roadside Picnic"

Khton

Fucking Ekibastuz

sample generation from AI Kandinsky

The first picture under the article title: Planet Plyuk, according to the ChatGPT neural network.

Tags: ,

Russian neural networks: their strengths and weaknesses






Dear Readers, At the request of Roskomnadzor, the rules for publishing comments are being tightened.

Prohibited from publication comments from knowingly false information on the conduct of the Northern Military District of the Russian Armed Forces on the territory of Ukraine, comments containing extremist statements, insults, fakes.

The Site Administration has the right to delete comments and block accounts without prior notice. Thank you for understanding!

Placing links to third-party resources prohibited!


  • November 2026
    Mon Tues Wed Thurs Fri Sat Total
    «    
     1
    2345678
    9101112131415
    16171819202122
    23242526272829
    30  
  • Subscribe to Politnavigator news