The team is assembled from AI models by different vendors. Here you can see which model is responsible for what shows where it excels, where it falls short, and how much its stage costs — so you choose not "the most expensive," but the right one.
The shortest explanation of who covers what: for each role — economical, balanced, and top-tier options. The model can be changed at any stage; the estimate will be recalculated before the run.
| Role | Economy | Balance | Top |
|---|---|---|---|
| Analyst собирает требования, пишет ТЗ, разбирает market and данные | or-nemotron-ultra Free workhorse model for the platform: handles specs, market analysis, architecture, and code, and a context window of up to 1M tokens lets you feed it an entire project. | groq-gpt-oss-120b Открытая model OpenAI 120B на сверхбыстром железе Groq: сильные рассуждения, code and структурированные ответы за секунды. Бесплатный тариф. | claude-opus-5 Самые сложные рассуждения: architecture системы, разбор противоречивых требований, длинные документы целиком. Берите там, где error в решении more expensive… |
| architect проектирует систему: стек, схему данных, интеграции | or-nemotron-ultra Free workhorse model for the platform: handles specs, market analysis, architecture, and code, and a context window of up to 1M tokens lets you feed it an entire project. | groq-gpt-oss-120b Открытая model OpenAI 120B на сверхбыстром железе Groq: сильные рассуждения, code and структурированные ответы за секунды. Бесплатный тариф. | claude-opus-5 Самые сложные рассуждения: architecture системы, разбор противоречивых требований, длинные документы целиком. Берите там, где error в решении more expensive… |
| Team lead ведёт команду and project, принимает решения на гейтах | or-nemotron-ultra Free workhorse model for the platform: handles specs, market analysis, architecture, and code, and a context window of up to 1M tokens lets you feed it an entire project. | groq-gpt-oss-120b Открытая model OpenAI 120B на сверхбыстром железе Groq: сильные рассуждения, code and структурированные ответы за секунды. Бесплатный тариф. | claude-opus-5 Самые сложные рассуждения: architecture системы, разбор противоречивых требований, длинные документы целиком. Берите там, где error в решении more expensive… |
| Product manager ценность, приоритеты, метрики продукта | or-nemotron-ultra Free workhorse model for the platform: handles specs, market analysis, architecture, and code, and a context window of up to 1M tokens lets you feed it an entire project. | claude-sonnet-5 Working лошадка команды: code, ТЗ, тексты, ревью. Quality близко к топовым моделям when цене в 2,5 раза ниже opus — model by default for большинств… | claude-opus-5 Самые сложные рассуждения: architecture системы, разбор противоречивых требований, длинные документы целиком. Берите там, где error в решении more expensive… |
| Developer пишет code, правит and рефакторит | or-nemotron-ultra Free workhorse model for the platform: handles specs, market analysis, architecture, and code, and a context window of up to 1M tokens lets you feed it an entire project. | groq-gpt-oss-120b Открытая model OpenAI 120B на сверхбыстром железе Groq: сильные рассуждения, code and структурированные ответы за секунды. Бесплатный тариф. | claude-opus-5 Самые сложные рассуждения: architecture системы, разбор противоречивых требований, длинные документы целиком. Берите там, где error в решении more expensive… |
| QA пишет тесты and проверяет результат stages | or-nemotron-ultra Free workhorse model for the platform: handles specs, market analysis, architecture, and code, and a context window of up to 1M tokens lets you feed it an entire project. | groq-gpt-oss-120b Открытая model OpenAI 120B на сверхбыстром железе Groq: сильные рассуждения, code and структурированные ответы за секунды. Бесплатный тариф. | — no models in this class |
| Technical writer documentation, инструкции, описания | or-nemotron-ultra Free workhorse model for the platform: handles specs, market analysis, architecture, and code, and a context window of up to 1M tokens lets you feed it an entire project. | groq-gpt-oss-120b Открытая model OpenAI 120B на сверхбыстром железе Groq: сильные рассуждения, code and структурированные ответы за секунды. Бесплатный тариф. | gemini-2.5-pro · on request Very long context: processes large documents, research papers, and legislation in full, without chunking. |
| Marketer позиционирование, тексты кампаний, конкуренты | or-nemotron-ultra Free workhorse model for the platform: handles specs, market analysis, architecture, and code, and a context window of up to 1M tokens lets you feed it an entire project. | groq-gpt-oss-120b Открытая model OpenAI 120B на сверхбыстром железе Groq: сильные рассуждения, code and структурированные ответы за секунды. Бесплатный тариф. | grok-4 · on request Reasoning and up-to-date topics, unconventional phrasing — useful when you need text that isn't market-average. |
| Content manager поток текстов: карточки, посты, рассылки | gemini-flash Fast Google model on the free tier: documentation, summaries, fact extraction, and bulk text processing. | — no models in this class | grok-4 · on request Reasoning and up-to-date topics, unconventional phrasing — useful when you need text that isn't market-average. |
| Lawyer договоры, оферты, соответствие требованиям | — no models in this class | mistral-large · on request European provider (data processed in the EU) and low price: short business texts and European languages. | claude-opus-5 Самые сложные рассуждения: architecture системы, разбор противоречивых требований, длинные документы целиком. Берите там, где error в решении more expensive… |
| Researcher обзоры, гипотезы, научная литература | or-nemotron-ultra Free workhorse model for the platform: handles specs, market analysis, architecture, and code, and a context window of up to 1M tokens lets you feed it an entire project. | deepseek-reasoner · on request Step-by-step reasoning and math at low cost: task decomposition, checking calculations, working through logic. | claude-opus-5 Самые сложные рассуждения: architecture системы, разбор противоречивых требований, длинные документы целиком. Берите там, где error в решении more expensive… |
| Translator перевод and локализация | groq-qwen3-27b Qwen3 27B на Groq: хороший code and Russian, мгновенные ответы. | qwen-max · on request Multilingual tasks, including Chinese; solid code and business writing, reasonable price for its class. | gemini-2.5-pro · on request Very long context: processes large documents, research papers, and legislation in full, without chunking. |
| Scriptwriter сценарии роликов, storyboard, тексты for video | or-nemotron-ultra Free workhorse model for the platform: handles specs, market analysis, architecture, and code, and a context window of up to 1M tokens lets you feed it an entire project. | claude-sonnet-5 Working лошадка команды: code, ТЗ, тексты, ревью. Quality близко к топовым моделям when цене в 2,5 раза ниже opus — model by default for большинств… | grok-4 · on request Reasoning and up-to-date topics, unconventional phrasing — useful when you need text that isn't market-average. |
| Circuit design engineer электроника, платы, схемы | — no models in this class | deepseek-reasoner · on request Step-by-step reasoning and math at low cost: task decomposition, checking calculations, working through logic. | claude-opus-5 Самые сложные рассуждения: architecture системы, разбор противоречивых требований, длинные документы целиком. Берите там, где error в решении more expensive… |
| Process chemist материалы, рецептуры, техпроцессы | — no models in this class | — no models in this class | claude-opus-5 Самые сложные рассуждения: architecture системы, разбор противоречивых требований, длинные документы целиком. Берите там, где error в решении more expensive… |
| копирайтер | glm-flash GLM-4.5-Flash from Z.ai: бесплатная, fast, уверенный Russian and китайский, good держит формат and структуру ответа. Mode размышлений отключён на шлю… | cf-llama-70b Llama 3.3 70B на Cloudflare Workers AI (fp8, fast): универсальные тексты, анализ, ТЗ, documentation, переписка. Бесплатно ~10 000 нейронов/день (≈… | — no models in this class |
| менеджер of the project | — no models in this class | cf-llama-70b Llama 3.3 70B на Cloudflare Workers AI (fp8, fast): универсальные тексты, анализ, ТЗ, documentation, переписка. Бесплатно ~10 000 нейронов/день (≈… | — no models in this class |
| технический писатель | glm-flash GLM-4.5-Flash from Z.ai: бесплатная, fast, уверенный Russian and китайский, good держит формат and структуру ответа. Mode размышлений отключён на шлю… | cf-llama-70b Llama 3.3 70B на Cloudflare Workers AI (fp8, fast): универсальные тексты, анализ, ТЗ, documentation, переписка. Бесплатно ~10 000 нейронов/день (≈… | — no models in this class |
| тестировщик | cf-qwen-coder Кодовая model Qwen 32B на Cloudflare Workers AI: vibe coding, генерация тестов, рефакторинг, скрипты. Бесплатно ~10 000 нейронов/день. | — no models in this class | — no models in this class |
| devops | cf-qwen-coder Кодовая model Qwen 32B на Cloudflare Workers AI: vibe coding, генерация тестов, рефакторинг, скрипты. Бесплатно ~10 000 нейронов/день. | — no models in this class | — no models in this class |
| ассистент | ministral-8b Ministral 8B — лёгкая fast model Mistral for простых задач and черновиков. Бесплатно, экономит лимиты. | — no models in this class | — no models in this class |
| редактор | ministral-8b Ministral 8B — лёгкая fast model Mistral for простых задач and черновиков. Бесплатно, экономит лимиты. | — no models in this class | — no models in this class |
| Role | What it does | Economy | Balance | Top |
|---|---|---|---|---|
| Product manager | Images and design | — no models in this class | openai-image Better than others at preserving the meaning of long prompts and edits like "leave it as is, just change this"; decent on-image text. | — no models in this class |
| Product manager | Video | — no models in this class | — no models in this class | veo-3 The best video in the registry and the only one with built-in sound — a finished scene, not a silent clip. |
| Developer | Images and design | openai-compatible-image · on request Any OpenAI-compatible /images/generations endpoint: your own Stable Diffusion, a corporate gateway, or a Russian reseller. | — no models in this class | — no models in this class |
| Technical writer | Voiceover | google-tts · on request The cheapest commercially clean voiceover in the registry, and the provider offers a free monthly quota (several million characters). | yandex-speechkit Russian voiceover directly from Russia, payment in rubles, live voices (alena, filipp and др.) — best choice for a Russian-language video; the provider yes ещё более д… | — no models in this class |
| Marketer | Images and design | alice Яндекс YandexART («Шедеврум», движок Алисы) через домашний браузер-мост — Running of РФ без зарубежных keys and без лимитов OpenAI, good понимает… | flux-pro-1.1 A reliable model for final raster frames: clean photorealistic images at a predictable price. | gemini-pro-image · on request Top-tier image quality (Nano Banana Pro): complex scenes, precise detail, 4K. |
| Marketer | Video | kling-2.5-turbo · on request Cheap, decent-quality 720P video with a choice of aspect ratio. | runway-gen4 Animating a still frame (image→video): take a ready-made image and get motion — a predictable result. | veo-3 The best video in the registry and the only one with built-in sound — a finished scene, not a silent clip. |
| Marketer | Music and sound | — no models in this class | elevenlabs-music Ready-made music track from a description — video background, jingle, voiceover bed. Cheap per second. | — no models in this class |
| Marketer | Voiceover | — no models in this class | yandex-speechkit Russian voiceover directly from Russia, payment in rubles, live voices (alena, filipp and др.) — best choice for a Russian-language video; the provider yes ещё более д… | elevenlabs-tts Best voice quality on the market: intonation, emotion, cloning — for ads and voiceovers customers actually hear. |
| Content manager | Images and design | alice Яндекс YandexART («Шедеврум», движок Алисы) через домашний браузер-мост — Running of РФ без зарубежных keys and без лимитов OpenAI, good понимает… | flux-pro-1.1 A reliable model for final raster frames: clean photorealistic images at a predictable price. | — no models in this class |
| Content manager | Video | minimax-hailuo-fast · on request The cheapest per-second video in the registry — for mass-producing short scenes. | — no models in this class | — no models in this class |
| Content manager | Voiceover | google-tts · on request The cheapest commercially clean voiceover in the registry, and the provider offers a free monthly quota (several million characters). | openai-tts Even, natural voice, six ready-made voices, a simple synchronous call — mid-range pricing. | — no models in this class |
| Designer | Images and design | alice Яндекс YandexART («Шедеврум», движок Алисы) через домашний браузер-мост — Running of РФ без зарубежных keys and без лимитов OpenAI, good понимает… | flux-pro-1.1 A reliable model for final raster frames: clean photorealistic images at a predictable price. | gemini-pro-image · on request Top-tier image quality (Nano Banana Pro): complex scenes, precise detail, 4K. |
| Designer | Video | — no models in this class | runway-gen4 Animating a still frame (image→video): take a ready-made image and get motion — a predictable result. | — no models in this class |
| Scriptwriter | Images and design | flux-2-klein The cheapest frame in the registry: storyboard drafts, composition variants, quick idea iteration. | — no models in this class | — no models in this class |
| Scriptwriter | Video | veo-3-lite · on request Google's cheapest video model — a moving rough storyboard for testing a video concept. | luma-ray Smooth camera movement and transitions; one of the cheapest options at 720p. | veo-3 The best video in the registry and the only one with built-in sound — a finished scene, not a silent clip. |
| Scriptwriter | Voiceover | — no models in this class | — no models in this class | elevenlabs-tts Best voice quality on the market: intonation, emotion, cloning — for ads and voiceovers customers actually hear. |
| editor | Video | minimax-hailuo-fast · on request The cheapest per-second video in the registry — for mass-producing short scenes. | runway-gen4 Animating a still frame (image→video): take a ready-made image and get motion — a predictable result. | veo-3 The best video in the registry and the only one with built-in sound — a finished scene, not a silent clip. |
| editor | Music and sound | — no models in this class | elevenlabs-music Ready-made music track from a description — video background, jingle, voiceover bed. Cheap per second. | — no models in this class |
| editor | Voiceover | — no models in this class | yandex-speechkit Russian voiceover directly from Russia, payment in rubles, live voices (alena, filipp and др.) — best choice for a Russian-language video; the provider yes ещё более д… | — no models in this class |
| composer | Music and sound | — no models in this class | elevenlabs-music Ready-made music track from a description — video background, jingle, voiceover bed. Cheap per second. | — no models in this class |
| narrator | Voiceover | google-tts · on request The cheapest commercially clean voiceover in the registry, and the provider offers a free monthly quota (several million characters). | yandex-speechkit Russian voiceover directly from Russia, payment in rubles, live voices (alena, filipp and др.) — best choice for a Russian-language video; the provider yes ещё более д… | elevenlabs-tts Best voice quality on the market: intonation, emotion, cloning — for ads and voiceovers customers actually hear. |
| Model | Strengths and suited tasks | When not to use | Roles | Stage |
|---|---|---|---|---|
| cf-llama-70b balance Cloudflare | Llama 3.3 70B на Cloudflare Workers AI (fp8, fast): универсальные тексты, анализ, ТЗ, documentation, переписка. Бесплатно ~10 000 нейронов/день (≈ сотни ответов). ТЗ, аналитика рынка, documentation, emails | Дневной limit нейронов; по коду and длинному контексту уступает Gemini and gpt-oss-120b. | Analyst, копирайтер, Marketer, менеджер of the project, технический писатель fast · Russian: good · 128K | 0 ₽ |
| gemini-flash economy Google | Fast Google model on the free tier: documentation, summaries, fact extraction, and bulk text processing. Write documentation, Summarization, Data extraction from text, Batch description writing | Free tier is capped by requests per minute and per day; gives shallow answers on complex reasoning and architecture. | Technical writer, Content manager, QA, Marketer fast · Russian: good · long window; check the exact value with the provider | 0 ₽ |
| glm-flash economy Z.ai | GLM-4.5-Flash from Z.ai: бесплатная, fast, уверенный Russian and китайский, good держит формат and структуру ответа. Mode размышлений отключён на шлюзе, чтобы не съедать limit вывода. тексты and переводы, описания and documentation, структурированные ответы, черновики | Слабее топовых models на длинных рассуждениях and сложном коде; платные GLM недоступны — на счёте ноль. | копирайтер, Translator, Analyst, технический писатель, Marketer fast · Russian: good · 128K | 0 ₽ |
| groq-gpt-oss-120b balance Groq | Открытая model OpenAI 120B на сверхбыстром железе Groq: сильные рассуждения, code and структурированные ответы за секунды. Бесплатный тариф. Write a spec, market analysis, architecture, code and рефакторинг, QA report | Limits бесплатного тарифа (запросов в минуту/день); на очень длинных документах уступает Gemini. | Analyst, architect, Developer, Marketer, Team lead fast · Russian: good · 128K | 0 ₽ |
| groq-gpt-oss-20b economy Groq | Лёгкая and очень fast: черновики, сводки, классификация. краткие сводки, черновики документации | Слабее 120B в рассуждениях and русском. | Technical writer, Content manager, QA fast · Russian: average · 128K | 0 ₽ |
| ministral-8b economy Mistral | Ministral 8B — лёгкая fast model Mistral for простых задач and черновиков. Бесплатно, экономит лимиты. черновики, короткие ответы, переводы | Мал по объёму знаний, for сложного не годится. | ассистент, редактор fast · Russian: good · 128K | 0 ₽ |
| mistral-small balance Mistral | Mistral Small — универсальная model: тексты, анализ, documentation, хороший Russian and европейские языки. Бесплатный тариф. тексты, анализ, documentation, emails | Слабее топовых на сложных рассуждениях; free-тариф ограничивает запросы в секунду. | Analyst, копирайтер, Marketer, технический писатель fast · Russian: good · 128K | 0 ₽ |
| or-nemotron-ultra economy OpenRouter · NVIDIA Nemotron | Free workhorse model for the platform: handles specs, market analysis, architecture, and code, and a context window of up to 1M tokens lets you feed it an entire project. Write a spec, market analysis, Design the architecture, Code refactoring, Write documentation, Large document analysis | Provider's free tier: request limits and queueing, responses can be slow; falls behind paid flagships on the hardest problems, and Russian phrasing sometimes needs editing. | Analyst, architect, Developer, QA, Technical writer average · Russian: average · up to 1M tokens | 0 ₽ |
| claude-haiku-4.5 economy Anthropic | Fast and cheap: documentation for existing code, short summaries, routine edits and format checks. Handles Russian well for its price. Write documentation, Shorten text, Check against a checklist, Bulk wording edits, Brief summary of the artifact | Not for complex reasoning or architecture: makes mistakes on extended logic and loses structure in long documents. | Technical writer, QA, Content manager, Translator fast · Russian: good · 200K tokens | 2 ₽ |
| claude-opus-5 top Anthropic | The most demanding reasoning: system architecture, resolving conflicting requirements, entire long documents. Use it where a wrong decision costs more than the stage's price. Design the architecture, Untangle contradictory requirements, Review a contract, Scientific review, solution and code audit, Write a complex spec | Expensive for high-volume tasks (2.5x sonnet's input price and 5x haiku's) and noticeably slower; overkill for documentation and routine text. | Analyst, architect, Team lead, Lawyer, Researcher slow · Russian: excellent · up to 1M tokens, no surcharge for long context | 8 ₽ |
| Model | Strengths and suited tasks | When not to use | Roles | Stage |
|---|---|---|---|---|
| cf-qwen-coder economy Cloudflare | Кодовая model Qwen 32B на Cloudflare Workers AI: vibe coding, генерация тестов, рефакторинг, скрипты. Бесплатно ~10 000 нейронов/день. code, тесты, скрипты, рефакторинг | Контекст 32K — большие репозитории не влезают; рассуждения слабее 120B-models. | Developer, тестировщик, devops fast · Russian: average · 32K | 0 ₽ |
| codestral economy Mistral | Codestral 22B from Mistral — специализированная кодовая model: генерация, автодополнение, тесты, рефакторинг на 80+ языках. Бесплатный тариф, отвечает быстро. code, тесты, рефакторинг, скрипты | Не for длинных рассуждений and не for общего of text; на free-тарифе limit запросов в секунду. | Developer, тестировщик, devops fast · Russian: average · 128K | 0 ₽ |
| groq-qwen3-27b economy Groq | Qwen3 27B на Groq: хороший code and Russian, мгновенные ответы. code, рефакторинг, перевод | Контекст and глубина ниже, чем у 120B. | Developer, QA, Translator fast · Russian: good · 128K | 0 ₽ |
| or-laguna-code economy OpenRouter | Free model for code: generation, editing, and source analysis with no token charges. Write code, Code refactoring, automated tests, Analyzing unfamiliar code | Free tier: request limits and queueing; noticeably weaker outside code (writing, analysis) — better to route documentation to another model. | Developer, QA average · Russian: average · Check with the provider | 0 ₽ |
| claude-sonnet-5 balance Anthropic | The team's workhorse: code, specs, writing, review. Quality close to top-tier models at 2.5x lower cost than opus — the default model for most roles. Write a spec, Generate a landing page, Code refactoring, Change review, Task work plan, Requirements analysis | Falls behind opus on contentious architecture and conflicting requirements; pricier than haiku and cheap Asian models for bulk routine work. | Developer, architect, Analyst, QA, Marketer average · Russian: excellent · up to 1M tokens, no surcharge for long context | 3 ₽ |
| Model | Strengths and suited tasks | When not to use | Roles | Stage |
|---|---|---|---|---|
| alice economy Alice | Яндекс YandexART («Шедеврум», движок Алисы) через домашний браузер-мост — Running of РФ без зарубежных keys and без лимитов OpenAI, good понимает Russian промпт. logo по описанию, illustration for a post, key frame, картинка на русском промпте | Идёт через headless-браузер владельца (сессия Алисы), поэтому медленнее прямого API and зависит from живости моста and кук; вектор doesn't allow (растровый JPEG). | Designer, Marketer, Content manager average · Russian: good | 0 ₽ |
| alice-art economy Alice | YandexART через Chat Алисы (мост на .251). Бесплатно, account владельца. Русские промпты — сильная сторона. | — | 0 ₽ | |
| cf-flux-schnell economy Cloudflare | Cloudflare Workers AI, free ~10k нейронов/день (≈50-100 картинок); FLUX.1 schnell, 4 шага | — | 0 ₽ | |
| flux-2-klein economy Black Forest Labs | The cheapest frame in the registry: storyboard drafts, composition variants, quick idea iteration. storyboard, draft frame variants, trying different compositions, illustrations for a draft | Detail is lower than the pro versions; text and labels in images come out poorly — don't use for final shots. | Designer, Scriptwriter, Marketer, Content manager fast · Russian: average | 2 ₽ |
| openai-image-mini economy OpenAI | Cheap OpenAI image model: understands complex prompts and edits well, good for illustration streams. illustration for a post, simple infographic, image from description, preview stream | In low mode quality drops noticeably; photorealism is weaker than flux-pro. | Content manager, Marketer, Designer fast · Russian: good | 2 ₽ |
| flux-pro-1.1 balance Black Forest Labs | A reliable model for final raster frames: clean photorealistic images at a predictable price. frame for the video, illustration for an article, banner, background for the landing page | Draws on-image text poorly; can't produce vectors (SVG) — for logos use Recraft Vector. | Designer, Marketer, Content manager average · Russian: average | 5 ₽ |
| openai-image balance OpenAI | Better than others at preserving the meaning of long prompts and edits like "leave it as is, just change this"; decent on-image text. key frame, cover mockup, image with text, a series of illustrations in one style | Price depends heavily on quality (low/medium/high differ by up to 35x) — at high the stage becomes expensive. | Designer, Marketer, Product manager average · Russian: good | 6 ₽ |
| recraft-v3-vector balance Recraft | Native SVG — the only way in the registry to get a true vector: logos, marks, icons, brand identity elements. logo, mark for the brand book, icon set, vector illustration | More expensive than raster models and not suited to photorealism; can't convert complex scenes to vector. | Designer, Marketer, Product manager average · Russian: average | 9 ₽ |
| Model | Strengths and suited tasks | When not to use | Roles | Stage |
|---|---|---|---|---|
| runway-gen4 balance Runway | Animating a still frame (image→video): take a ready-made image and get motion — a predictable result. animate a finished frame, logo animation, motion in the banner | Requires a starting frame (params.image_url), doesn't generate from text alone; the key must have the key_ prefix and an API version header. | editor, Designer, Marketer average · Russian: average | 6 ₽ |
| luma-ray balance Luma | Smooth camera movement and transitions; one of the cheapest options at 720p. camera passes, atmospheric scene, transition between frames | ⚠ per the price list, 1080p is abnormally expensive — check your rate in your Luma account; the field with the finished-file link has not been verified. | editor, Scriptwriter average · Russian: average | 7 ₽ |
| cogvideox-3 economy Z.ai | CogVideoX-3: текст→video and кадр→video, до 4K, 30/60 fps, with sound. $0.2 за video. Needs ненулевой balance Z.ai. | — | 23 ₽ | |
| veo-3 top Google Veo | The best video in the registry and the only one with built-in sound — a finished scene, not a silent clip. final scene of the video, video with sound, ad frame, presentation video | The most expensive entry in the registry: eight seconds of video costs as much as several hundred text stages. Minimum 4 s, 8 s for 1080p and 4K. | editor, Scriptwriter, Marketer, Product manager slow · Russian: average | 46 ₽ |
| Model | Strengths and suited tasks | When not to use | Roles | Stage |
|---|---|---|---|---|
| elevenlabs-music balance ElevenLabs | Ready-made music track from a description — video background, jingle, voiceover bed. Cheap per second. background music for the video, jingle, background track for voiceover, soundtrack for a presentation | ⛔ ElevenLabs officially does not serve Russia or Belarus; commercial use is available only on paid plans. | composer, editor, Marketer average · Russian: average | 2 ₽ |
| elevenlabs-sfx balance ElevenLabs | Sound effects from a description: impacts, UI sounds, ambience. Fixed price per generation. transition sound, interface sounds, ambient noise, accent in the video | You pay per generation, not per second — short effects end up expensive; ⛔ not served to Russia by the provider. | editor, composer fast · Russian: average | 4 ₽ |
| Model | Strengths and suited tasks | When not to use | Roles | Stage |
|---|---|---|---|---|
| openai-tts balance OpenAI | Even, natural voice, six ready-made voices, a simple synchronous call — mid-range pricing. English voiceover, audio version of the text, voice for a demo | A Russian accent is noticeable; from Russia — only via proxy. | narrator, Content manager fast · Russian: good | 2 ₽ |
| yandex-speechkit balance Yandex SpeechKit | Russian voiceover directly from Russia, payment in rubles, live voices (alena, filipp, etc.) — the best choice for a Russian-language video; the provider also has a cheaper API v3. Russian voiceover for the video, audio version of the article, voice for an answering machine, narration text | Requires the Yandex Cloud folder's folderId in the line settings; the choice of voices is more limited for foreign languages. | narrator, Marketer, editor, Technical writer fast · Russian: excellent | 2 ₽ |
| elevenlabs-tts top ElevenLabs | Best voice quality on the market: intonation, emotion, cloning — for ads and voiceovers customers actually hear. ad voiceover, brand voice, audiobook, emotional narration | The most expensive voiceover in the registry — dozens of times pricier than Google TTS; ⛔ ElevenLabs officially does not serve Russia or Belarus. | narrator, Marketer, Scriptwriter average · Russian: good | 21 ₽ |
| Model | Strengths and suited tasks | When not to use | Roles | Stage |
|---|---|---|---|---|
| ollama-qwen2.5 economy Local (Ollama) | Runs on your own hardware: data never leaves your infrastructure, and tokens are free. A sound choice for sensitive documents and experiments with no API bill. Draft based on internal documents, Personal data processed on your own server, Offline experiments, Log analysis | Quality is noticeably below cloud models, and speed is limited by your GPU; better not to trust it with complex stages. | Developer, Technical writer, Analyst average · Russian: average · depends on the model build and available VRAM | 0 ₽ |
| ollama-qwen7b economy locally (Ollama .84) | Локальная 7B-model: данные не покидают контур, токены бесплатны; заметно грамотнее 3B-версии. черновики документации, краткие сводки, проверка текстов | Медленная на CPU (десятки секунд на ответ), слабее облачных на сложных рассуждениях. | Technical writer, QA, Content manager slow · Russian: good · 32K | 0 ₽ |
Stage price — benchmark for типового этапа (6000→1500 tokens) at the current rate and exchange rate; how it's calculated — see the page Plans. With your own provider key or your own Claude Max subscription, you pay only for orchestration. Model descriptions are maintained by the service administrator and updated in line with vendor pricing.