AI-COMFORTModelCatalog
Built-in AI-COMFORT models with transparent ruble billing plus the freedom to connect your own OpenAI-compatible models through BYOK.
Current CBR rate: 1 USD = 73.47 ₽
GPT-OSS-120b (OpenAI)
Reliable text model for everyday business work with large volumes of information. • Capabilities: Text only • Best suited for: Agreements, regulations, instructions, internal memos, long reports and other extensive texts. • Strengths: Provides even, clear and predictable responses, maintains long context well, useful for proofreading, shortening and rewriting materials in a business style. • Practical benefit: Can not only explain but also prepare a basis for precise calculation via a Python script, if the task is undesirable to do in one's head. • Limitations: Does not work with images and scans; for visual materials it is better to choose models with vision capabilities. • When to choose: If the main work is done through documents and you need a stable text assistant without unnecessary creative improvisation. • Context: 131K / 131K tokens.
DeepSeek V4 Flash (DeepSeek)
A fast universal model for everyday employee work. • Capabilities: Text‑only; • Best suited for: emails, short reports, summaries, service responses, processing large documents, internal material search; • Strengths: high speed, large context window, good cost‑quality ratio for mass corporate use; • Practical benefit: works not only with text but also with work artifacts — example, it can quickly build a Mermaid diagram of a process, approval stages, roles, or interaction logic; • Limitations: in especially complex analytical tasks it may fall short of stronger models; • When to choose: as the primary fast option “for everyday use” for most employees; • Context: 1.05M / 393K tokens.
Hy3 preview (Tencent)
An advanced text model for unconventional and more demanding work tasks. • Capabilities: Text only • Best suited for: Complex phrasing, ambiguous documents, analyses of contested situations, structuring unorganized information. • Strengths: Handles multi-step logic and tasks that require more than just answering, but careful unpacking. • Practical benefit: Useful for turning complex descriptions into clear structures — e.g., process schemes, decision trees, or formal action sequences. • Limitations: As a preview model, results can be less stable in some scenarios compared to more mature solutions. • When to choose: When the task goes beyond ordinary correspondence and requires thoughtful processing. • Context: 262K / 262K tokens.
👁️🗨 Gemma 4 31B (Google)
A balanced model for working with both text and visual materials. Capabilities: text + vision. It is best suited for: screenshots, photos of documents, scans, tables in images, as well as typical business text tasks. Strengths: assists with mixed material where part of the information is in text and part is on an image. Practical benefit: can extract data from visual material and turn it into a clear output — structured text, tables, diagrams, or a basis for calculations. Limitations: in the most critical tasks it may fall short of top closed models. When to choose: when an employee works with scans, screenshots, photos, and visual documents. Context: 131K / 131K tokens.
👁️🗨️ GPT-5.4 Nano (OpenAI)
A very fast multimodal model for routine work operations. • Capabilities: Text + Vision. • Best suited for: short answers, screenshot processing, data extraction from documents, email drafts, quick summaries and typical employee tasks. • Strengths: High speed, universality and convenience for mass use in daily work. • Practical benefit: Suitable for simple calculations via Python scripts, basic Mermaid diagrams and other applied results when a fast answer is needed. • Limitations: For deep analytics, complex documents and especially critical tasks, it's better to choose more powerful models. • When to choose: When a fast and practical result is needed without extra costs. • Context: 400K / 128K tokens.
👁️🗨️ Gemini 3.1 Flash Lite (Google)
An efficient multimodal model for streaming processing of work materials. • Capabilities: Text + Vision • Best suited for: File processing, extracting data from scans, quick summaries, translations, material classification, and routine office queries. • Strengths: Works fast, is low‑cost, and fits tasks that are repeated many times a day. • Practical benefit: Can quickly prepare a table, a template, a simple calculation script, or a basic Mermaid diagram from a task description. • Limitations: Not ideal for deep analytics or cases requiring the strongest possible final text. • When to choose it: For large‑scale daily processing of documents, images, and internal materials. • Context: 1.05M / 65K tokens.
DeepSeek V4 Pro (DeepSeek)
A powerful model for complex business tasks where depth and result quality matter. • Capabilities: Text only • Best suited for: Detailed analytical notes, comparative materials, regulations, complex reports, and high‑stakes textual tasks. • Strengths: Compared to faster models, it retains intricate reasoning structures and helps produce more cohesive, argument‑backed outputs. • Practical benefit: Especially useful when the answer must become a working tool — such as a complex Python calculation script, a business logic model, or a Mermaid process diagram. • Limitations: Slower and more expensive than lightweight models; often overkill for simple routine work. • When to choose: When depth, accuracy, and richness are more important than speed. • Context: 1.05M / 393K tokens.
GLM 5.1 (Z.ai)
A model for long, multi-step work tasks. • Capabilities: Text only • Best suited for: Large analytical tasks, step-by-step processing of complex questions, long documents and sequential preparation of materials "from draft to result". • Strengths: Handles long task chains well and helps maintain logic across large volumes of information. • Practical benefit: Ideal for scenarios where you need to sequentially reach a result — for example, first break down the task, then gather structure, and finally prepare a calculation or diagram. • Limitations: May be excessive for simple, short queries. • When to choose: When you need not just an answer, but to go through a long workflow together with the model to a final result. • Context: 202K / 65K tokens.
👁️🗨️ GPT-5.4 Mini (OpenAI)
A strong universal model for most serious work tasks. • Capabilities: Text + Vision • Best suited for: Document and screenshot analysis, preparing letters and reports, OCR scenarios, presentation materials and complex work queries. • Strengths: One of the best universal options in terms of quality, speed and breadth of application. • Practical benefit: It works not only for communication but also for creating work artifacts — precise calculations via Python scripts, Mermaid diagrams, structures, templates and rich business materials. • Limitations: More expensive than simple fast models, so it is not always optimal for the most mass routine workload. • When to choose: As the primary universal option for employees and managers when a strong and reliable result is needed. • Context: 400K / 128K tokens.
🖼️ Nano Banana 2 (Google)
Advanced Google graphics model for generating and editing images from textual descriptions. Capabilities: Image generation + image editing. Best suited for: marketing creatives, product cards, banners, presentation illustrations, visual concepts, covers, mockup images, and refining existing pictures. Strengths: It well understands the task context, supports step‑by‑step image refinement in a dialogue, and outperforms regular generators on applied visual materials where meaning, composition, and detail are crucial. Practical benefit: It helps not only to create an image from scratch but also to tweak it to fit a work task — change style, composition, details, background, product presentation, or visual emphasis. Limitations: This is a preview model, so the result may require several iterations; it is not designed for ordinary text answers, document analysis, or calculations. When to choose it: When you need to create or edit an image with a more precise understanding of the task than simple quick graphic models provide.
🖼️ Riverflow V2 Fast (Sourceful)
A fast graphic model for generating and editing working images. • Capabilities: Image generation + image editing • Best suited for: Marketing visuals, product cards, presentation illustrations, banners, packaging concepts, social media creatives, and quick visual drafts. • Strengths: High speed, good quality for production tasks, confident handling of composition, subject scenes, and applied visual materials. • Practical benefit: Helps you quickly obtain not just a “picture” but working visual material — e.g., a cover, ad creative, product mockup, document illustration, or a base for a presentation. • Limitations: It is not intended for text responses and document analysis; it does not support 4K output, so for the largest final images it's better to use other options. • When to choose: When you need to quickly create or edit an image for a business, marketing, or presentation task.
🖼️ Grok Imagine Image Quality (xAI)
Качественная графическая модель xAI для фотореалистичных изображений, рекламы и визуальных материалов с текстом. • Возможности: Генерация изображений + редактирование изображений • Лучше всего подходит для: Рекламных креативов, постеров, упаковки, меню, карточек товаров, социальных сетей, презентационных иллюстраций и реалистичных сцен с продуктами. • Сильные стороны: Хорошо передаёт реалистичный свет, материалы, текстуры и предметные сцены; лучше многих графических моделей справляется с читаемым текстом внутри изображения и брендированной визуальной подачей. • Практическая польза: Помогает создавать готовые маркетинговые материалы — обложки, баннеры, продуктовые mockup-изображения, постеры, меню, рекламные сцены и варианты визуала по референсу. • Формат изображения: Соотношение сторон задаётся прямо в тексте запроса, например: “квадрат 1:1”, “вертикальный формат 9:16”, “горизонтальный баннер 16:9”. • Ограничения: Не предназначена для обычных текстовых ответов, расчётов и анализа документов; для аккуратной доработки сложного изображения может потребоваться несколько итераций. • Когда выбирать: Когда нужен более качественный и реалистичный визуал, особенно для рекламы, упаковки, карточек товаров или изображений с текстовыми элементами.
Icons in model names
Vision (OCR) — the ability to recognize and analyze images: photos, screenshots, documents, diagrams.
This page shows the built-in AI-COMFORT catalog with ready-to-use settings, ruble billing, and business paperwork.
If your team needs other models, your own OpenRouter quotas, GigaChat, YandexGPT, or a self-hosted model on your own infrastructure, they can be connected separately through an OpenAI-compatible API.
Connected external models do not remain raw API models: AI-COMFORT applies the corporate system prompt, reasoning, temperature 0.3, tools, and other work-ready settings to them.
How tokens are billed
The minimum charge is 0.01 ₽ (1 kopeck). Applies separately to input and output tokens.
Pricingspecifics
Input tokens are always cheaper than output tokens.
Reading and analyzing text requires fewer computational resources than generating new responses.
Input tokens include corporate settings.
Beyond your prompts, the model receives about 3000–4000 additional tokens with platform corporate settings.
Output tokens depend on reasoning depth.
Models are set to medium reasoning level — an optimal balance between quality and token consumption.
Different models convert text to different token counts.
Better trained models consume fewer tokens for the same text.
Corporate Model Settings
All models in AI-COMFORT are configured for professional use: reasoning — medium, temperature — 0.3. This ensures accuracy and predictability with minimum hallucinations.
All models are 'thinking' (reasoning: medium) — best quality
Temperature 0.3 — minimum fabrications and hallucinations
Third-party services: temperature 0.8-1.0 — high error risk
Accuracy matters more than 'creativity' for business
Subscription + pay-as-you-go
This page shows the built-in AI-COMFORT catalog with transparent pricing and ruble billing. If your team needs other models, free quotas from your own provider, or your own infrastructure, they can be connected separately through an OpenAI-compatible API — and the platform applies the same corporate work profile to them.
View pricing plansStartusingAItoday
Deploy your work platform in 24 hours and give your team access to the best AI models.