590+ models · one balance · no subscription
First, look at what people make here
Real work by our users, settings on show. See something you like? Press “Repeat” and upload a photo of your own. The price appears before you run it, and a failed generation costs nothing.
What it actually costs
Five subscriptions, or one balance
A normal stack for someone who works with AI seriously. The money is only half of it: every subscription is another tab, another seat, another renewal date — and you pay for the seat whether you used it this month or not.
Buying the subscriptions directly
- ChatGPT Plus $20
- Claude Pro $20
- Midjourney Standard $30
- Suno Pro $10
- Kling Standard $8.80
- Five tabs and five passwords —
- Priced per user licence, not per use —
- A separate limit inside each service —
- Didn’t use it — you pay anyway —
NeuralBox
- All 590+ models included
- Subscription none
- One balance for everything yes
- Pay for what you use, not per user licence yes
- Balance never expires never
- Failed generation — refunded automatic
- Didn’t use it $0
Your actual cost depends on what you generate: a page of text costs cents, a minute of video costs dollars. The price of every model and every mode is shown before you run it. See all plans and per-model pricing
What a top-up buys
Priced in finished results, not in tokens
Pick an amount and see how many generations that is. Nothing to convert in your head — and before each run the exact cost of that operation is shown on its own.
Chat models are not on this list on purpose: text is billed by the tokens a conversation actually uses, so there is no fixed price per page to count in.
How much cheaper
The same models, below the official price
We buy access in bulk and re-price against the market. The saving differs by category, and we would rather show it as it is than average it into a single number.
| What you generate | Official vendor API price | Our price | Cheap mode price | You save |
|---|---|---|---|---|
| Nano Banana | $0.039 | $0.020 | $0.010 | up to −74 % |
| Seedream 5 Pro | $0.075 | $0.035 | $0.018 | up to −77 % |
| Nano Banana Pro · 1K | $0.134 | $0.090 | $0.045 | up to −66 % |
| Kling 3.0 · 1 s | $0.168 | $0.070 | — | up to −58 % |
| Veo 3 Lite · 8 s | $0.400 | $0.150 | $0.075 | up to −81 % |
On the left is the vendor’s own published API price for the same generation; on the right, ours at the $100 package — 5,000,000 tokens, or 50,000 tokens per dollar. The cheapest image in the catalog is $0.004.
Cheap mode — the same models for less
Some cards carry a switch: same model, same result, a little longer to wait. Most of them drop to half price. The GPT Image family also has a detail level, and that moves the saving both ways — 75% off at the lowest detail on GPT Image 2, 10% off at the highest on GPT Image 2.5. The form always shows the exact price.
| Model | Standard price | Cheap mode price | You save |
|---|---|---|---|
| Kling Image v3 · 1K | $0.012 | $0.006 | −50 % |
| Nano Banana | $0.020 | $0.010 | −50 % |
| Nano Banana 2 Lite | $0.020 | $0.010 | −50 % |
| Seedream 4.0 | $0.025 | $0.013 | −50 % |
| GPT Image 2.5 · 1K | $0.030 | $0.015 | −50 % |
| Nano Banana Pro · 1K & 2K | $0.090 | $0.045 | −50 % |
| Veo 3 Lite 720p · clip | $0.150 | $0.075 | −50 % |
Prices are shown for the $100 top-up — 5,000,000 tokens. The larger the package, the cheaper every operation.
Model access
Every model is available at any balance
No hunting for the tier that happens to include the model you need, and no upgrade to unlock it. Top up, and the whole catalog is open — from the cheapest model to the flagship.
| NeuralBox | Typical aggregator | Vendor subscription | |
|---|---|---|---|
| Whole catalog from the start | yes | by tier | one family |
| Commercial use | yes | paid plans only | paid plans only |
| No watermarks | yes | paid plans only | paid plans only |
| Your work stays private | yes | paid plans only | varies |
| Queue without paid priority | yes | priority costs extra | priority costs extra |
| Failed generation — refunded | automatic | not everywhere | rarely |
| Balance expiry | never | monthly | monthly |
| Top up without a subscription | yes | needs an active plan | subscription only |
| Training on your data | no | varies | varies |
Ready-made workflows
Not “access to models” — a finished result
The model, the settings and the chain are already chosen. You do not need to write a prompt — just upload a file and pick what to do with it.
Product photo → video
Upload a product shot, get a vertical clip for marketplaces and stories.
photo → scene → motion → 9:16
Bring an old photo to life
Restoration, colorization and gentle motion for an archive shot. The face is preserved.
scan → restore → color → video
Translate video with voice
Your clip in another language, keeping the voice, the lips and the timing. 190 languages.
video → translate → voice → lip sync
Your own model, on your photos
LoRA training: 15–20 shots, then any frame with your face, your product or your style.
photos → training → your own model
Remove what shouldn’t be there
Watermarks, objects and backgrounds — in photos and in video. Upscale to 4K.
file → mask → result
Result as a file
A quote with costings, a report with charts, a structured presentation — as a ready Word, Excel or PDF file.
task → draft → file
Community
Like it? Repeat it in two clicks
Pieces made by our users, with their prompts and settings on show. Press “Repeat”, upload a photo of your own, and get the same result built around your photo.
Catalog
590+ models — and you don’t have to choose
378 for text, 96 for images, 78 for video, 24 for music and 16 for voice. Each model says what it is best at. Would you rather not compare them yourself? Describe the task and we’ll pick for you.
Cinematic scenes with sound that follow the prompt closely.
Brings photos to life with smooth motion, multi-shot with one character.
Keeps faces and composition when editing. Accurate text in the image.
Follows instructions best and renders legible text in the frame.
The cheapest one, for trying ideas and testing prompts.
Long documents and deep analysis. A one-million-token context.
An all-round model for writing, code and working with files.
A full song with vocals from a description of genre and mood.
Expressive voiceover and cloning of your own voice.
What’s new
New models every week
Dates are when the model appeared on NeuralBox. The list is assembled from the catalog history automatically — nobody maintains it by hand.
- NEW TemPolor v4.8 and i4 — songs with vocals and pure instrumentals
- NEW Video text removal — removes watermarks, subtitles and on-screen text
- NEW GLM 5.3 Prime and Qwen3.8 Max Prime — top chat models from Zhipu and Alibaba
- Claude Opus 5.5 and GPT-6 Luna Pro — new Anthropic and OpenAI flagships for hard tasks
- Higgsfield Cinema Studio 4.0 — cinematic video from Higgsfield
- Genjutsu Motion Transfer — transfers motion from a video onto your photo
- Grok 4.7 — the new xAI chat model
- Qwen Image 2.1 — image generation and editing, up to 10 references
- ElevenLabs v3 Conversational — natural conversational voiceover with emotion
- Stem separation — vocals, bass, drums and backing track from any song
- FLUX 3 Video — video from text and photos, extend and upscale
- Grok 4.6 — chat with web search and file analysis
- GPT Image 2.5 — Flare and Sunburst — two quality tiers, up to 16 references
- Suno V6, V6 Mini and V6 Wild — the new generation of the music model
- GPT-6 Astra — OpenAI flagship for complex tasks and code
Developers and companies
One API key, one balance, one invoice
Integrate it through the API
Compatible with the OpenAI and Anthropic SDKs: change the base_url and it works. All 590+ models on one API key.
base_url="https://neuralbox.ru/api/v1",
api_key="sk-…",
)
- Works with Claude Code, Cursor, n8n and LangChain
- A budget and limits for each API key
- The cost of each API request comes back in the response and is written to the log
- A failed request is not billed
For your company
One balance for the team, roles and per-person limits, proper documents and a spend report you can read.
- Invoice, delivery note and contract for companies, with electronic document exchange
- Spending dashboard: who used which model and how much it cost
- A spending limit for each employee
- Your data is not used to train models — written into the contract
- Personal data does not go to third parties beyond what the generation itself requires
A provider failed — you didn’t notice
Every popular model has several execution routes. If one channel does not answer, the request goes to another channel running the same model.
Switching never costs more
We charge the price of the most expensive route, so a fallback channel can never cost you more than the amount you were shown.
We spot a provider failure before you do
Channel failures are tracked automatically: usually we are already fixing it while you are still deciding whether to write to support.
Wherever you are
Pay the way you already pay
Visa and Mastercard, Apple Pay and Google Pay, PayPal and Link. Card details are handled by the payment provider — we never store them.
Alipay for China, and USDT if you would rather not use a card at all.
Top up with Telegram Stars in our bot. The balance is the same one the website uses.
Top up once and spend it whenever. The balance never expires, and a month you skip costs $0.
Email link, Google, Apple, Microsoft, GitHub, Telegram, LINE or TikTok. No password to invent or forget.
Telegram and email. Answered by people, not a bot.
Fewer subscriptions. More work actually finished.
Signing up takes ten seconds and top-ups start at $5. The balance never expires, and a failed generation gives your money back.