YandexGPT in 2026: A Review of Russia's AI Platform for Business

12 min read
Stanislav Belyaev
Stanislav Belyaev Engineering Leader at Microsoft
YandexGPT in 2026: A Review of Russia's AI Platform for Business

YandexGPT is a family of generative language models from Yandex, available through Alice, Yandex Browser, the Yandex Cloud API, and other ecosystem services. By February 2026, Yandex introduced the new Alice AI model family, which together with YandexGPT forms a powerful platform for working with the Russian language.

Update: Alice AI Flash, the New Alice, and Benchmark Results (July 2026)

Alice AI LLM Flash (May 28, 2026) is Yandex’s new lightweight model. According to Yandex’s internal tests, it outperforms GPT-5.4 mini in 56% of tasks. It’s positioned as a fast, cheap option for high-volume requests – Yandex’s answer to Google’s Gemini Flash. Not yet tested in our benchmark.

Alice got an update (June 25, 2026) – cross-conversation memory (Alice now remembers context between sessions), an adaptive communication style, and improved document handling. According to Mediascope, Alice is the #1 AI assistant in Russia, used by 14.3% of all AI users.

Results from the MySummit benchmark (42 models, 80 management scenarios) place Yandex’s models as follows:

  • Alice AI LLM v3 – in the bottom quarter of the ranking, though noticeably better than v2. It costs $0.8/$0.8 per 1M tokens – the same as Claude-tier models – yet Gemma 4 26B beats it at a third of the price.
  • Alice AI LLM v2 – near the bottom of the ranking. There’s no reason to pick the base version when v3 is available.

The judges’ verdict: Alice AI v3 works for scripting difficult conversations with employees (high empathy), for Yandex’s corporate customers, and for internal analytical drafts. It doesn’t work for employment law (critical errors), financial modeling, automation, or code.

The only real justification for choosing Alice AI is an existing corporate contract with Yandex. Otherwise, DeepSeek V4 Flash is accessible from Russia, free, and clearly outscores Alice.

DeepSeek on Yandex Cloud. Yandex now offers DeepSeek V4 Flash through its own API – the best model in Yandex’s ecosystem, well ahead of Alice AI v3. But it costs $3/$5 per 1M tokens – 28-33 times more – than going directly to DeepSeek ($0.09/$0.18). You’re paying for the convenience of keeping data inside Russia’s regulatory perimeter, not for quality.

Цена vs Качество: где модель на карте рынка

Каждая точка – модель из BenchLM. Выше – сильнее, левее – дешевле. Выделены модели из этого обзора.


The key advantage of these tools is their deep understanding of the Russian language, cultural context, local realities, and legislation. The models were trained on Russian-language data and handle tasks better where post-Soviet specifics, knowledge of Russian legislation, or business practices in Russia are important.

Using YandexGPT in Alice
Using YandexGPT in the Alice chatbot

YandexGPT Models (February 2026)

In August 2025, Yandex released access to its latest model, YandexGPT 5.1 Pro, which outperforms the previous version in 58% of cases and OpenAI’s GPT-4.1 in 56% of cases (according to Yandex).

YandexGPT compared to competitors
According to Yandex, YandexGPT outperforms ChatGPT. There is no independent confirmation of this yet.

YandexGPT 5.1 Pro

The flagship model, available in Yandex Cloud:

  • 128,000 token context window (approximately 240 pages of text) – a significant increase from 32K in the previous generation.
  • Inference is 2.5 times faster compared to YandexGPT 4 Pro.
  • Supports Chain-of-Reasoning mode – similar to reasoning models (o3, DeepSeek R1).
  • Ranks second in the SLAVA benchmark (a comprehensive benchmark for evaluating LLMs in Russian) after Alice AI LLM.

Alice AI LLM

Yandex Alice AI LLM (v3) Needs supervision score 6.3 #43 of 47 $0.002 per task available in Russia
tested: July 2026 · mysummit.school benchmark About the model → Compare →

The next-generation core language model, the leader of the SLAVA benchmark for Russian. Optimized for dialogue, context extraction, and personal assistant tasks.

Alice AI LLM and YandexGPT are different model families. Alice AI was designed for agentic/assistant scenarios (Alice), while YandexGPT targets a broader range of tasks via API.

YandexGPT Lite

Lightweight models (versions 4 and 5) for quick responses and simple tasks. They run faster and are cheaper, but with limited capabilities. Ideal for chatbots, quick queries, and bulk operations.

Shedevrum: Image and Video Generation

Shedevrum is Yandex’s neural network for creating images and videos (up to 6 seconds in 16:9, 9:16, and 1:1 formats).

Shedevrum – image generation
Yandex Shedevrum: an example of a prompt-generated image

Original from Shedevrum
Original image generated by Shedevrum

Ecosystem Capabilities

  • Access through Alice – use YandexGPT via the voice assistant on smartphones, smart speakers, and other devices.
  • Yandex Browser integration – a built-in AI assistant for working with web page content, text summarization, and answering questions about page content.
  • Yandex 360 for business – integration with corporate services (email, cloud storage, documents, boards). Alice Pro analyzes reports, generates text, and searches files by context. Yandex Documents supports up to 400 simultaneous users.
  • API for developers – Yandex Cloud provides an API for integrating YandexGPT into custom applications.
  • Working with files and images – document and visual content analysis in the Pro version.

At first glance, Yandex’s ecosystem looks self-sufficient – models, APIs, integrations with everyday work tools. But in practice, model choice determines the quality of the outcome, and prompting mistakes with YandexGPT often go unnoticed until you compare the output against alternatives.

In the free module – 9 tasks that show you how AI handles real manager work. Free.

No payment required • Get notified on launch

Join Waitlist

Alice Pro: Corporate AI Assistant

Alice Pro is an enhanced version of the AI assistant, integrated with data from your Yandex 360 subscription. In terms of functionality, it resembles Google NotebookLM – allowing you to work with corporate documents and data.

Alice Pro interface
The Yandex Alice Pro interface

Services Built on YandexGPT

Yandex focuses not so much on top-tier models as on creating “useful services” for users:

  • Alice chatbot and smart speakers – a voice assistant for everyday tasks. Since June 2026, it has cross-conversation memory (context carried between sessions) and an adaptive communication style. The #1 AI assistant in Russia (14.3% of AI users, per Mediascope).
  • Shedevrum – image generation and coloring pages for children.
  • Video transcription service300.ya.ru.
  • Built-in video transcription and summarization in Yandex Browser.
  • Alice Pro – a Google Workspace-like AI solution – lets you work with data from your subscription.

YandexGPT Pricing

YandexGPT – Free or Paid?

The basic version is free. YandexGPT is available through Alice and Yandex Browser without a subscription. Since July 2025, most of the advanced features in “Chat with Alice” built on the Pro model – reasoning mode and file handling – have also been free.

The corporate option is Alice Pro through a Yandex 360 for business subscription. Pricing depends on headcount and the service bundle.

YandexGPT API Pricing (as of 2026)

For developers and automation – through Yandex Cloud and AI Studio (starting March 3, 2026):

ModelCost per 1M tokensQuality tier
YandexGPT Lite₽200Simple tasks, drafts
YandexGPT Pro 5.1₽800Main workhorse model
Alice AI LLM FlashTBDNew lightweight model (May 2026), fast tasks
Alice AI LLM v3₽500 input / ₽1,200 outputBottom quarter of the benchmark
DeepSeek V4 Flash (Yandex)~₽270 input / ~₽450 outputBest model in Yandex’s ecosystem
DeepSeek V4 Flash (direct)~₽8 input / ~₽16 outputStrong and very cheap
GigaChat 3.5 Ultra~₽900Bottom of the ranking

Alice AI LLM v3 costs 60 times more than DeepSeek V4 Flash (direct) – despite scoring lower in the benchmark. DeepSeek V4 Flash through Yandex Cloud is more convenient (data stays in Russia), but costs 28-33 times more than direct access for slightly worse quality.

Free trial API access is available through starter grants and promo codes in Yandex Cloud.

You've just seen how many models and price tiers there are – but which one actually fits your tasks? In the free module, we walk through tool selection using 9 real-world cases.

No payment required • Get notified on launch

Join Waitlist

Practical Applications

  • Content creation for Russian audiences with local specifics in mind.
  • Document analysis in Russian – contracts, reports, presentations.
  • Working with legislation – Russian regulatory acts and legal documents.
  • Corporate workflow integration through Yandex 360.
  • Chatbots and virtual assistants for Russian-speaking users.

Try It Yourself: A Workplace Conversation Scenario

Try it yourself
Русский деловой сценарий – YandexGPT vs DeepSeek vs GigaChat
You
You're a department head at a mid-sized tech company in Austin, Texas (120 employees). Draft a conversation script for an employee who has been showing up 30-40 minutes late every day for the past month and is unreachable during the first hour of the workday. ## Context - The employee has been with the company for 2 years, no prior issues - The team has noticed the lateness, and two people have already asked "why does he get away with it" - You suspect personal problems (a divorce) but nothing is confirmed - Work format: hybrid, 3 office days required per week ## Task 1. Structure of the conversation: what to say in the opening, the main body, and the close 2. Specific phrases – empathetic but direct (no corporate jargon) 3. How to respond to 3 possible reactions: "I'm dealing with stuff at home," "I don't see a problem here," silence 4. What commitments to document afterward 5. What to do if nothing changes in 2 weeks – with references to relevant U.S. employment law
Comparing:
aliceai-llm-v3 · deepseek-v4-flash · GigaChat-2-Max

What to watch for: Alice AI (YandexGPT) usually produces reasonably empathetic, well-structured phrasing for the conversation itself, even in English. But when it comes to citing U.S. employment law (point 5), it frequently gets things wrong – inventing statutes or misapplying provisions that don’t exist. DeepSeek is cheaper and more consistent on the analytical parts of the task, while GigaChat holds up decently given it wasn’t built for English-language business writing.

Coming Soon

YandexGPT Is Just One of the Tools

In the course, you'll compare YandexGPT, ChatGPT, Claude, and DeepSeek on real management tasks. 9 free lessons: from your first working prompt to a framework for choosing the right model. No coding, built for managers.

In-depth tool breakdowns with real examples
Ready-to-use prompts for common tasks
Safe and responsible AI usage skills
How to measure and communicate AI ROI
Start for free →
No payment required

Continue learning

Open the textbook and pick up where you left off

Open Textbook

Results in Our Benchmark

In the MySummit benchmark, we tested 42 AI models across 80 real management scenarios – from project planning to financial report analysis. All prompts were written in Russian, from the perspective of an ordinary manager.

Current Results (July 2026)

Alice AI LLM (v3) across 8 task categories

0–10 scale. Higher means the model handled those tasks better

chart

Alice AI LLM v3 – an improvement over v2, but still in the bottom quarter. Works for scripting employee conversations (high empathy). Critical errors in employment law and financial modeling.

Alice AI LLM v2 – no reason to use it now that v3 exists.

YandexGPT Pro 5.1 – showed weak results in the benchmark’s earlier waves. In half of the information-lookup scenarios, the model refused the task outright, citing a “lack of current data” – even for well-established, widely known topics.

DeepSeek V4 Flash on Yandex Cloud – the best model in Yandex’s ecosystem, well ahead of Alice AI v3. But it costs 28-33 times more than going directly to DeepSeek ($3/$5 vs. $0.09/$0.18 per 1M tokens). You’re paying for the Russian data perimeter, not for quality.

The regional-specificity paradox. On tasks with Russian and Kazakh context, Russian models don’t outperform Chinese alternatives. Kimi K2.6, MiMo v2.5 Pro, and DeepSeek V4 – all accessible from Russia without a VPN – score noticeably higher. A model’s analytical ability matters more than its “native” language.

Full interactive results →

Limitations

  • Quality trails global leaders – Alice AI LLM v3 is Yandex’s best in-house model (per our benchmark), but it falls well behind Chinese alternatives available from Russia (Kimi K2.6, MiMo v2.5 Pro, DeepSeek V4 Pro). Prompting can close part of the gap – we tested ten techniques across thousands of queries and found which ones help and which make things worse.
  • Loses even on Russia-specific tasks – Chinese models (DeepSeek V4, Kimi K2.6, MiMo v2.5) outperform Alice in the “regional specificity” category.
  • Ecosystem lock-in – full capabilities are unlocked only within the Yandex ecosystem.
  • Limited context – 128K tokens vs 1M for Claude and 2M for Gemini.
  • No global community – fewer integrations, plugins, and third-party tools.
Stanislav Belyaev

Stanislav Belyaev

Engineering Leader at Microsoft

18 years leading engineering teams. Founder of mysummit.school. 700+ graduates at Yandex Practicum and Stratoplan.