Skip to main content

General

NeuralBox API gives developers access to 52 AI models from 9 providers through a single REST API. Instead of managing separate accounts with OpenAI, Anthropic, Google, MidJourney, ElevenLabs and others — you get one endpoint, one token balance, and a unified response format.
No. NeuralBox handles all provider integrations. You only need a NeuralBox account and API key.
Partially. The request format for text generation is OpenAI-compatible (same messages array format). However, the base URL and authentication header are different. See the Quickstart for details.

Billing & Tokens

NeuralBox tokens are internal credits, not LLM context tokens. One NeuralBox token represents a fixed compute budget. Each model has a fixed cost in NeuralBox tokens per request. See the Tokens page for the full pricing table.
No. Subscription tokens reset each billing cycle. However, token packages (one-time purchases) never expire and carry over indefinitely.
Requests return 402 Payment Required. You can top up with a token package or upgrade your subscription from the Dashboard.
The Starter plan gives you access to Starter-tier models (19 models) and 50 one-time tokens. All paid plans (Basic and above) unlock all 52 models. Plans differ only in monthly token allowance — not model access.

Technical

Long-running generations return a generation ID immediately with status: pending. Poll GET /api/v2/generate/{id} until the status changes to completed. See the Video Generation Guide for a full example.
Rate limits depend on your plan. See the Rate Limits page for details.
Yes. NeuralBox is designed for the Russian market. All AI provider calls are routed through a Cloudflare Workers proxy, so you don’t need a VPN. Payments are accepted via Yookassa (Russian cards, SBP) and Telegram Stars.
Official Python and JavaScript SDKs are in development. In the meantime, the API is standard REST — any HTTP client works. See the SDKs page.

Models

Start with the Models Catalog for a full list with token costs. For most tasks: gpt-5-nano or claude-haiku-4.5 for text (0 tkn), dall-e-3 for images (5–13 tkn), flux-1.1-pro for photorealistic images (7 tkn), kling-v2.1 for video (26–94 tkn).
New models are added as providers release them, typically within days of a major release. Check the Changelog for updates.
Yes. NeuralBox uses the LegNext proxy to provide MidJourney access via API — no Discord bot required. Use midjourney-v6 or midjourney-turbo for images, and mj-video-480p / mj-video-720p for videos.