Best of the week

More neural networks on Bothub

Claude Sonnet 4.6

A universal “speed + intelligence” model, optimized for everyday development, agentic programming, and automation—from writing and editing code to working with tools and long contexts. Claude Sonnet 4.6 is often chosen as the “workhorse” for teams: it helps get tasks done faster, follows instructions better, and is well-suited for large-scale professional scenarios (including 1M tokens in beta via API).

Nano Banana 2

Nano Banana 2 (Gemini 3.1 Flash Image) is a Google image generation model with Search Grounding for accurate 4K outputs. Produces legible text in any language, keeps up to 5 consistent characters, and is 2× faster than the Pro version.

Veo 3.1

Veo 3.1 from Google DeepMind is a video generator up to 4K with native sound and dialogue. Creates 8-second clips with realistic physics, frame control and reference images in 16:9 and 9:16 formats.

GPT-5.2

GPT-5.2 is an OpenAI language model with three modes (Instant, Thinking, Pro) and a 400K-token context window. It delivers 38% fewer errors, improved performance on code and STEM tasks, and a knowledge cutoff in August 2025.

Flux-2

Flux-2 from Black Forest Labs is a photorealistic image generator up to 4K with support for 10 reference images at once. Features precise typography, 32B parameters and open Dev weights, and ranks first on the Image Arena benchmark.

Gemini 3.1 Pro Preview

Advanced multimodal model from Google with early access to the next generation. Natively works with text, images and audio — it «sees», «hears» and analyzes content. A universal tool for developers and business.

models-page.best-models.link

Available neural network models

Client type

Individual

Business

Tariff

ELITE

Product

APIDashboard

Currency

RUBCAPS

			Cost in rubles
Model	Context size (in tokens)	Output size (in tokens)	Prompt (per 1M tokens)	Image prompt (per 1k tokens)	Response (per 1M tokens)
gpt-5.4	1 050 000	128 000	294,64	0	1 767,86
gpt-5.4-pro	1 050 000	128 000	3 535,71	0	21 214,29
gpt-5	400 000	128 000	147,32	0	1 178,57
gpt-5.2	400 000	128 000	206,25	0	1 650
o3-deep-research	4 095	100 000	1 178,57	0	4 714,29
gpt-5.3-codex	400 000	128 000	206,25	0	1 650
gpt-4.1	1 047 576	345 700	235,71	0	942,86
gpt-5.5	1 050 000	128 000	589,29	0	3 535,71

* Our markup on these prices is 5%, which is included in the cost of packages except Basic (Premium and higher)

LLM Request

Cost of a single request in the dashboard

All tariffs

Used tokens + 0.01 $per 1 request

Special attention: The use of Easy Writer is charged differently. For each text generation, Easy Writer charges an additional 0.1 $ per request + the token cost as specified above for a regular LLM request.

Image Generation

Cost of a single generation by models

MidJourney — Relax

26 000 CAPS / 4,09 ₽ For 1 generation

MidJourney — Fast

52 000 CAPS / 8,17 ₽ For 1 generation

MidJourney — Turbo

104 000 CAPS / 16,34 ₽ For 1 generation

Dall-E

33 000 CAPS / 5,19 ₽ For 1 generation

Flux

1 666 CAPS / 0,26 ₽ For 1 generation

Stable Diffusion

39 375 CAPS / 6,19 ₽ For 1 generation

GPT Image - Square

8 160 CAPS / 1,28 ₽ For 1 generation

GPT Image - Portrait

12 240 CAPS / 1,92 ₽ For 1 generation

GPT Image - Landscape

12 000 CAPS / 1,89 ₽ For 1 generation

Nano Banana — Pro

100 800 CAPS / 15,84 ₽ For 1 generation

Nano Banana — 2

112 000 CAPS / 17,6 ₽ For 1 generation

Nano Banana

29 025 CAPS / 4,56 ₽ For 1 generation

Web Search

Cost of a single web search usage

All tariffs

Used tokens + 0.01 $per 1 request

Link Analysis

0,01 Capsfor 1 character

Video Generation

Cost of creating one second of video

GoogleVeo

168 750 Caps / 26.52 ₽per 1 second

Runway

30 000 Caps / 4.71 ₽per 1 second

Sora

337 500 Caps / 53.04 ₽per 1 second

Kwaivgi

189 000 Caps / 29.70 ₽per 1 second

For video generation in 1080p quality using veo-3, an additional charge of +20% is added

Speech Synthesis

Cost of one speech synthesis

TTS

11 250 Caps / 1.77 ₽per 1 000 characters

TTS HD

27 225 Caps / 4.28 ₽per 1 000 characters

Transcription

The cost of one transcription

AssemblyAI — nano

2 000 Caps / 0.314 ₽Per 1 minute

AssemblyAI — best

5 500 Caps / 0.864 ₽Per 1 minute

A fixed surcharge on all requests: $0.05 per request, $0.10 for files over 50 MB, $0.50 for files over 500 MB

Embeddings

Model embeddings available through our API.

		Cost in Caps	Cost in dollars
Model	Embedding dimension	Prompt cost (per 1 token)	Prompt cost (per 100,000 tokens)
text-embedding-3-largeThe most efficient embedding model	3 072	0,12	0,16
text-embedding-3-smallIncreased performance compared to the 2nd generation ada embedding model	1 536	0,02	0,02
text-embedding-ada-002The most powerful 2nd generation embedding model, replacing 16 first generation models	1 536	0,09	0,12

text-embedding-3-largeThe most efficient embedding model

3 072Embedding dimension

0,12Prompt cost (per 100,000 tokens)

0,16Prompt cost (per 100,000 tokens)

text-embedding-3-smallIncreased performance compared to the 2nd generation ada embedding model

1 536Embedding dimension

0,02Prompt cost (per 100,000 tokens)

text-embedding-ada-002The most powerful 2nd generation embedding model, replacing 16 first generation models

1 536Embedding dimension

0,09Prompt cost (per 100,000 tokens)

0,12Prompt cost (per 100,000 tokens)

What are Caps?

Caps is the internal currency of the service, used to measure the cost of requests and responses of neural networks. It is fixed and depends on the model complexity: number of parameters, multimodality, and overall power.

For example:

ChatGPT-3.5 — ~1 Caps per token

ChatGPT o1-Pro — ~400+ Caps per token

The higher your tariff, the better the price: 1 million Caps is cheaper on Elite than on Basic.

Caps to dollar exchange rate for corporate clients

Still have questions?

What are tokens?

Tokens are units of text processing by the neural network, representing parts of words, entire words, or punctuation marks that determine the cost of requests.

How long will 1 million tokens last?

One million tokens of the GPT-4o model are enough to rewrite “The Brothers Karamazov” by F. M. Dostoevsky.

What to do if I run out of tokens?

Purchase additional Caps in your personal account — https://bothub.chat/profile

Why does the neural network pretend to be another?

The neural network does not know what model it is if it is not specified in the system prompt. The “self-identification” of the model without instruction is influenced by many factors, one of them being the model's data training set.

What is context in a neural network?

Context is the amount of information that the neural network retains in memory during a dialogue, affecting the coherence of responses and understanding of previous requests.

What is the context of different neural network models?

GPT o1 Pro and Claude 3.7 Sonnet support up to 200K tokens, Gemini 2.5 Pro works with 1KK, while Gemini 2.0 Pro supports up to 2KK tokens.

What file formats do models read?

Neural networks process TXT, PDF, DOCX, XLSX, CSV, JSON, XML, HTML, as well as images JPG, PNG, and audio files MP3, MP4.

Can neural networks be used for free?

There are free models with the postfix “:free” and “-exp” that can be used for free through a mini-window on the main page, as well as the model page.

How do neural network models differ from each other?

Models differ in the volume of training data, context size, processing speed, specialization in specific tasks, and ability to work with multimodal content.

How to use models via API?

To integrate models into your applications, you need to obtain an API key in your personal account. More details can be found here: https://bothub.chat/api/documentation/ru.

Can neural networks be used to automate business processes?

Neural networks effectively automate routine tasks of document management, data processing, customer support, and analytics, integrating with existing business systems via API.

Support ServiceOpen from 07:00 AM to 12:00 PM