Chat, images and video,through one API

NezhaGate is an OpenAI-compatible AI model gateway that connects you to GPT, Claude, Gemini and image-generation models through one API and one key. Instead of integrating each provider separately, you point your Base URL at NezhaGate and call multiple flagship models with the OpenAI SDK you already use. The platform includes API key management, balance and usage tracking, a model market and developer docs, billed by usage with no charge on failed requests.

NezhaGate currently provides the GPT-5.6 family (Sol / Terra / Luna), GPT-5.5 and GPT Image 2, Claude Sonnet 4.6, Gemini 3.1 Pro and Gemini 3.6 / 3.5 / 3 / 2.5 Flash, plus Nano Banana image generation and Veo 3.1 video, with new models added over time. The full, live model list and capability notes are shown on the Model Market page. Every model is called through the same endpoint and key.

NezhaGate uses prepaid credits and pay-as-you-go billing: chat models are billed by input / output tokens and image models by count and resolution, with unit prices shown on the Pricing page and Model Market. There are no monthly plans — you pay for what you use, and the balance is drawn down by actual usage. Every call's usage and cost is visible in the console.

Failed requests are not charged. Only calls that return successfully are billed by actual usage; upstream errors or timeouts do not consume your balance. If you see a failed record in the console, no cost is deducted for it.

Getting started takes three steps: register an account, create an API key in the console, then point your Base URL at NezhaGate and add the key. Because it is fully OpenAI-compatible, existing OpenAI SDK code needs almost no changes — just swap base_url and api_key. See the developer docs for details.

An API key is shown in full exactly once, at creation; afterwards the console shows only its prefix, and you can disable or delete it at any time. Each key carries its own spend cap, requests-per-minute limit, model allowlist and IP allowlist, so a leaked key never reaches your other projects. Every call and every balance change is itemised, so your bill always reconciles with your usage. Request content is used to fulfil the call and bill it -- never to train models.