GigaChat in 2026: Honest Review – Is It Worth Using for Work?
AI models in this article
GigaChat is a generative AI model from Sber – Russia’s largest bank and technology group. Its defining feature is that it’s a genuinely Russian-language model, trained on Russian data with a deep understanding of cultural context, slang, and the realities of the Russian language. And it works without a VPN.
For a manager, this means the model understands Russian business specifics – from corporate etiquette to legal nuances – better than competitors. In 2025, Sber released GigaChat 2.0, an updated generation with multimodal capabilities, and in March 2026 launched GigaChat Ultra, a new flagship model with a reasoning mode.
Update: GigaChat 3.5 Ultra and Benchmark Results (July 2026)
Sber released GigaChat 3.5 Ultra, the newest model in the lineup. It’s a mixture-of-experts model with 432 billion parameters – the largest in Sber’s history. Sber open-sourced the model weights on HuggingFace and GitVerse under the MIT License. Currently, for online testing within the Sber ecosystem, the model is only available via the web chat (giga.chat), mobile apps, and Telegram bot; GigaChat 3.5 Ultra is not yet available in the official Sber Cloud API (developers.sber.ru). Sber claims 3.5 Ultra outperforms GPT-4o on Russian-language tasks, but in our benchmark it lands in the bottom half of the ranking.
We tested it in the MySummit benchmark (42 models, 80 management-task scenarios). Here’s where every GigaChat model landed:
| Model | Cost per 1M tokens | Ranking position |
|---|---|---|
| GigaChat 3.5 Ultra | $10/$10 | Best in the lineup, lower half |
| GigaChat Ultra | $10/$10 | Near the very bottom |
| GigaChat 2 Max | $7.22/$7.22 | Closes out the ranking |
GigaChat 3.5 Ultra costs as much as GPT-4 but stays in the bottom half of the ranking. It’s surprisingly strong in strategic analysis – the best result among all Russian models in that category. Its weakest category is team management.
Benchmark judges’ verdict: suitable for letters to executives, meeting minutes, strategic analysis of alternatives, and situations that require direct access from Russia. Not suitable for financial calculations, references to employment law (dangerous mistakes), or team-management tasks.
For comparison: Gemma 4 31B delivers comparable results at 26 times lower cost. DeepSeek V4 Flash is free and scores noticeably higher. GigaChat’s only real advantage is guaranteed data storage inside Russia.
There is progress: from GigaChat 2 Max to 3.5 Ultra is a noticeable jump in scores. But the leaders (Claude Opus 4.6, MiMo v2.5 Pro from Xiaomi) are still far ahead.
Full access to GigaChat (model selection, all features) requires a SberID account and a Russian phone number. Basic functions are available without registration on the site, as well as through bots on Telegram and VK.
Try It Yourself: GigaChat vs DeepSeek vs Qwen
Before we get into the models and pricing, run a prompt right here and compare three models: GigaChat-2-Max (Sber), DeepSeek V3.2, and Qwen 3.5 Plus (Alibaba). These are previous-generation models, kept here for comparison – DeepSeek and Qwen have since released V4 and 3.6/3.7 (see the updated benchmark results above). But even on these older versions, the gap with GigaChat is visible.
Example 1. Internal communication – hybrid work announcement
This task tests writing fluency, a feel for business etiquette, and the ability to sound like a human being rather than a bureaucratic template at the same time. GigaChat is playing on its home turf here – Russian business vocabulary and cultural context. DeepSeek and Qwen are interesting as a test: do they hold up to the local tone, or does the output read like a translation? Later in the article there’s a second prompt on an analytical task, where the ranking shifts.
GigaChat Models
GigaChat 2.0 Lineup
- GigaChat 2 Lite – for fast everyday tasks: answering questions, drafting emails, simple summarization. Works quickly, minimal cost. Suited for chatbots and bulk request processing.
- GigaChat 2 Pro – balance of speed and quality. Handles summarization, rewriting, text editing, and creative tasks well (marketing copy, social media content). Better at following complex instructions.
- GigaChat 2 Max – a powerful model for complex professional tasks: analytics, working with large documents (up to 200 pages at once), deep data analysis. Context window: 128,000 tokens.
GigaChat Ultra (March 2026)
In March 2026, Sber launched GigaChat Ultra – a new flagship model with two modes:
- GigaChat Ultra – standard mode. The best GigaChat model according to our benchmark, though it placed in the lower group of the ranking. Structures responses well with tables and specific recommendations, but regularly makes factual errors. The model shows a chain of reasoning before answering. In our test across 32 management scenarios, the reasoning mode worsened the result – it helped on multi-factor tasks (stakeholder analysis, negotiation prep) but hurt where precise data and calculations were needed. A detailed breakdown with data is in a separate article.
GigaChat Ultra is available for free at giga.chat, in mobile apps, and via the Telegram bot. At the time of this review, the API was not yet available.
GigaChat Models Ranked in Our Benchmark
In our benchmark we tested every available GigaChat model on 80 real management tasks (42 models in the ranking). Every GigaChat model is in the bottom half, though the newest 3.5 Ultra has closed the gap noticeably.
GigaChat 3.5 Ultra – the best result in the lineup and a serious step forward. Surprisingly strong in strategic analysis, but team management is one of its worst categories. It costs $10/M tokens – like GPT-4, not like a model from the bottom third of the ranking.
GigaChat Ultra – the former flagship, now near the very bottom of the ranking. The most expensive mediocrity in the benchmark: it loses to Gemini Flash Lite ($0.15/M) and even to Alice AI ($0.8/M).
GigaChat Ultra Thinking – despite the reasoning mode, it scored slightly below standard Ultra. Paradoxically, communication is one of its worst categories, even though it’s the best category for Ultra without thinking. Learning and development is very weak.
GigaChat 2 Max – closes out the ranking. Alice AI costs 9 times less and scores noticeably higher.
A separate paradox is regional expertise. On its “home turf” (Russian legislation, Kazakhstani tax rules), GigaChat Ultra Thinking was flagged as a source of potential hallucinations: it mixed up Kazakhstani government agencies and cited incorrect tax rates. The Chinese model Kimi K2.6 handles comparable scenarios significantly better.
Meanwhile, GigaChat 3.5 Ultra’s price ($10/M) is comparable to Claude Sonnet 4.6, which scores noticeably higher. The price-to-quality ratio does not favor GigaChat.
Multimodal Capabilities of GigaChat 2.0
- Documents – upload a report, contract, or regulatory document (up to 200 pages in the Max version) and ask questions about the content. The model finds the relevant clause, generates a summary, or compares two documents.
- Audio – upload a meeting or interview recording (up to 1 hour) – GigaChat transcribes it, highlights key points, and compiles minutes. No prior conversion to text needed.
- Video – send a video link – the model describes the content and answers questions about it.
- Images – generate illustrations from descriptions (via the built-in Kandinsky image model) and analyze uploaded images: charts, diagrams, document photos.
- Music – create background compositions up to 3 minutes long for presentations or videos.
- Web search – finds current information and provides source links, similar to Perplexity.
- Voice calls – you can call GigaChat and discuss a task by voice: the model responds in real time, supports interruptions, and saves a transcript of the conversation.
- Deep Research – the model independently conducts multi-step analysis on complex questions, gathering information from multiple sources and forming a structured response with citations.
- Reasoning mode – breaks complex tasks into steps and shows a chain of reasoning, as if an expert were solving the problem. Useful for checking the logic behind an answer.
- Russian legislation – GigaChat navigates Russian laws, GOST standards, and regulatory documents better than competitors.
The feature list looks impressive – but there’s a gap between “the model can analyze documents” and “the model gives the right answer to your question.” Which tasks GigaChat handles reliably, and where you need to double-check, depends on how you phrase your request.
GigaChat understands Russian better than competitors – but the quality of the answer depends on the prompt. The free module has 9 tasks that show the difference between a good and a bad request.
No payment required • Get notified on launch
Choosing between GigaChat, ChatGPT, and DeepSeek is half the job. The other half is learning to get the result you actually want from a model. Free module: 9 management tasks with the mistakes broken down.
No payment required • Get notified on launch
Pricing
GigaChat – free or paid?
Basic access is free. The web interface at giga.chat works after signing in with SberID – no subscription and no daily message limits. GigaChat is also available for free on Telegram, VK, and Sber’s MAX messenger.
For the API (automation, product integration), Sber gives 1,000,000 free tokens per year – enough for several months of active use by one person.
GigaChat API Pricing (current for 2026)
In February 2026, Sber updated its pricing. Here’s how it compares to the closest competitors – models that also work from Russia without a VPN:
| Model | Cost per 1M tokens | Quality tier |
|---|---|---|
| GigaChat 2 Lite | 65₽ | Simple requests, drafts |
| GigaChat 2 Pro | 500₽ | Mid-tier texts and analysis |
| GigaChat 2 Max | 650₽ | Complex analytics |
| DeepSeek V4 Flash | ~$0.09/M | Strong and very cheap (benchmark) |
| Qwen 3.6 Plus | ~30₽ | Upper part of the ranking |
| ChatGPT (GPT-4o) | ~750₽ | Elite group |
| Claude Sonnet | ~900₽ | Elite group |
GigaChat 3.5 Ultra costs $10/M – like GPT-4 and Claude Sonnet. DeepSeek V4 Flash ($0.09/M) is 110 times cheaper with higher quality. If VPN-free access is your main criterion, DeepSeek, Qwen, and MiMo all work directly from Russia and cost less.
What GigaChat Costs in Practice: Our Experience
We use the GigaChat API on the MySummit.school learning platform – students run prompts directly from the articles and compare answers across models. Here are the real numbers for February–May 2026:
- 928 API requests (mostly GigaChat Max)
- 976,000 tokens used (~1,000 tokens per request)
- 3,450₽ spent – two 3-million-token packages (Max + Pro)
- Of the 6 million purchased tokens, less than 1 million had been used by publication time – the balance continues to be drawn down
An average request to GigaChat Max costs 0.68₽. The same 928 requests through DeepSeek would have cost 24₽ instead of 634₽ – a 26x difference, even though DeepSeek scores significantly higher in our benchmark.
The main trap is the package model. The GigaChat API sells tokens in 3-million-token packages, with a minimum of 1,500–1,950₽. Even with active use through the learning platform and hundreds of requests, we’ve used only 16% of the purchased volume. Competitors (DeepSeek, Qwen, OpenAI) charge pay-as-you-go – exactly what you used, nothing more.
Practical Applications for Managers
- Documents and reports – summarizing long documents, comparing contract versions, preparing executive summaries. GigaChat understands Russian business vocabulary better than foreign alternatives.
- Legislation and regulation – questions about Russian laws, GOST standards, and regulatory documents. The model is trained on Russian legal data.
- Marketing and content – texts for Russian audiences, social media posts, marketing materials with an understanding of local context and tone.
- Meetings – upload an audio recording of a meeting (up to 1 hour) and get a transcript with key points and a list of decisions.
- Presentations – GigaChat can generate a presentation from a text description.
If you already have GigaChat and want to get the most out of it, see our separate breakdown of prompt techniques that give a 20–30% quality boost, and the pitfalls of the reasoning mode: GigaChat for work – how to get the most out of what you already have.
Example 2. Analytical task – compliance and risk
The second prompt is analytical, testing the area where GigaChat should have an edge: knowledge of Russian regulatory frameworks. The models in this prompt are of the previous generation (V3.2, Qwen 3.5), kept here for comparison.
Pay attention to two things: how well the model handles GDPR-specific terminology (lawful basis, data minimisation, DPIA, consent vs. legitimate interest) and how tightly it holds onto every requirement in the prompt – 6 to 8 items, explanations, 2 flagged problem areas. That’s the difference between a pretty-looking answer and a working tool for a manager.
Limitations and Honest Comparison
In our benchmark, GigaChat 3.5 Ultra is the best result in the lineup, but it’s still in the bottom half of the ranking. The gap with the leaders remains large: Kimi K2.6, DeepSeek V4 Pro, and MiMo v2.5 Pro are all available from Russia and significantly stronger.
In practice, this means:
- For analytics and complex reasoning – ChatGPT, Claude, Kimi, and DeepSeek are significantly stronger
- For Russian-language texts, legislation, and business correspondence – GigaChat 3.5 Ultra is competitive, though Yandex’s Alice AI LLM v3 is also improving
- For multilingual tasks – ChatGPT and Claude hold the response language more consistently with mixed-language prompts
- Reasoning mode (Thinking) – worsens results in five of eight task categories. Detailed breakdown with data and links to academic research
The context window for all GigaChat 2.0 models is 128,000 tokens (comparable to GPT-4o and Claude Sonnet).
Meanwhile, GigaChat’s API pricing remains among the highest of Russian models. DeepSeek V4 Flash costs $0.09 per million tokens – 110 times cheaper with significantly higher quality.
Advantages of GigaChat
- Best understanding of Russian – trained on Russian-language data, understands business slang, cultural context, and local realities
- Russian legislation – navigates laws, GOST standards, and regulatory documents better than foreign alternatives
- All in one – text, images, audio, video, music, presentations, web search
- Availability – free web interface, Telegram bot, MAX messenger. API with a free annual limit
- Sber ecosystem – built into the MAX messenger, ATMs, and SberBoom smart speakers
- Data stored in Russia – data storage within the Russian Federation, compliant with 152-FZ. Important for companies with data localization requirements
GigaChat in Depth: Pros and Cons by the Data
Since this review, we’ve published three separate deep-dives – on the benchmark, on the reasoning mode, and on real-world use. Here’s what they add up to.
Where GigaChat is genuinely strong:
- Closed-loop deployment. It’s the only Russian model with full on-premise deployment: data never leaves the company perimeter. According to Sber, 65% of corporate contracts run this way. For banks, defense companies, and the public sector, this settles the choice. The SIBUR and Severstal cases are in the practical guide.
- GOST-compliant formatting. Requisites, formal register, business correspondence – GigaChat breaks format less often than foreign models here.
- Responsiveness to prompting. A well-built prompt gives a 20–30% quality boost – concrete techniques are broken down in the same guide.
Where it loses:
- Analysis and accuracy. When Sber published 40 business case studies, we checked them against a 29-model benchmark (first wave, May 2026 – the ranking now has 42 models). The flagship Ultra placed 28th, and the public 2 Max (which most of the cases actually run on) placed 29th, last. The cases prove the tool works, not that it’s optimal.
- Reasoning mode. With Thinking on, results get worse in five of eight categories – especially where precise numbers matter. Turn it off for calculation tasks.
- Price. The API stays among the most expensive of Russian models, while DeepSeek and Qwen are cheaper and stronger on analysis.
Bottom line for a manager: GigaChat gets chosen for the closed-loop deployment and Russian-document work. The model’s own quality is secondary here – without those requirements, competitors give more for less money.
GigaChat is one tool among many. Learn how to pick the right one
In the course, you'll test GigaChat, ChatGPT, Claude, and YandexGPT on the same management tasks – and understand when to use which. 9 free lessons, no coding required. Your first result comes in lesson one.
Useful Links
Stanislav Belyaev
Engineering Leader at Microsoft18 years leading engineering teams. Founder of mysummit.school. 700+ graduates at Yandex Practicum and Stratoplan.