What's the difference between Chat and Platform?

Chat is an app you use in the browser — pick a model, start a conversation, switch models mid-thread without losing context. Platform is the API: one key and one endpoint that reaches every model, for putting them inside your own product. They're separate products with separate pricing, and you can use either or both.

Do I need to know how to code?

Not for Chat — it works like any other chat app, with nothing to install or configure. Platform is for developers building on the API, and assumes you're comfortable making HTTP requests or using an SDK.

Which models can I use?

Frontier models from Anthropic, OpenAI, xAI, DeepSeek, Moonshot, Z.AI, MiniMax and ByteDance, plus de-risked variants you won't find on the original providers. Every model is reachable through the same endpoint and the same request format — you switch by changing the model name.

What does de-risked mean?

A de-risked model has had its refusal layer removed. The weights and capability are unchanged — what's gone is the classifier that intercepts a prompt and decides you shouldn't get an answer. It matters for work that mainstream models decline despite being legitimate: security research, clinical detail, mature fiction, frank financial analysis.

Do I have to rewrite my code?

No. The API uses the standard chat-completions format, so in most cases you change the base URL and the model name and your existing client keeps working — including streaming, tool calls, and structured output. Common SDKs work unmodified.

Will a model change under my integration?

No. Models are addressed by pinned identifiers, so a new version is a new name rather than a silent swap. Deprecations are announced ahead of time, and nothing changes behaviour underneath a running integration without notice.

Do you train on my data?

No. Prompts and completions sent through the API are not used to train or fine-tune any model, and they are not stored — requests are processed transiently. We keep only operational data such as token counts, timestamps, and error codes, which is what billing and reliability depend on.

Where do I find pricing?

Chat and Platform are priced separately — Chat by plan, Platform by token — so each has its own pricing page rather than one combined table. Both are linked from the product you're signing up to.

20 September 2026

Vani Chat is live, with plans from $18 a month

Vani Chat is open to everyone, plans start at $18 a month on one shared allowance, and every model now has a public status page. Here is what shipped and what comes next.

A white sun-shaped emblem glowing in a dark studio under amber light

Three things shipped this week: Vani Chat, subscription plans, and a public status page. The API is unchanged and still half price.

Vani Chat is open

Vani Chat puts every model in our catalogue behind one conversation: Claude, GPT, Grok, Gemini, DeepSeek, Kimi, GLM and the rest. Pick a model, start talking, and switch to another one mid-thread without losing your history or explaining yourself again. There is nothing to install.

That includes Claude Fable 5.1 and GPT-6 Astra, the two highest-scoring models on the Artificial Analysis Intelligence Index when we checked on 19 September 2026, and Claude Sonnet 5 for fast everyday work.

You can try it free at vani.ai.

Plans on one shared allowance

Every plan includes every model. What changes between plans is how much you can use them.

Free, $0: every model, at a fifth of Plus capacity, with 50 MB of storage.
Plus, $18 a month: 5× the usage of Free, extra-high and max reasoning effort, and 2 GB of storage.
Pro, $78 a month: 4.8× the usage of Plus, and 10 GB of storage.
Max, $188 a month: 2.4× the usage of Pro, and 50 GB of storage.
Ultra, $464 a month: 2.5× the usage of Max, and 100 GB of storage.

A plan is one allowance, not a quota per model. Spend it on a heavyweight when the problem is hard and on a lighter model when you need volume. On Plus, the same allowance is roughly 31 million tokens of Claude Fable 5.1, or 33 million of GPT-6 Astra, or 125 million of Claude Sonnet 5, or any mix of them. Pacing is weekly, and your weekly and monthly usage is tracked in one place.

Those figures are estimates for the whole allowance spent on a single model, counting input, cached input, repeated conversation context and output. Prices are in USD, billed monthly through Stripe, and you can change or cancel at any time. Full details are on the pricing page.

A public status page

status.vani.ai is live. Checks run continuously against each model, so you can see what is up before you start, and what happened if something was not. Times are shown in UTC.

The API is still 50% off

If you are building rather than chatting, platform.vani.ai serves 45 models from 13 providers through one endpoint, with 30 frontier models at half the official price. API usage is pay per token and billed separately from plans.

What comes next

Image and video generation in chat. We are finishing this now.
Mobile apps.
Your plan as an API. Soon you will be able to use your subscription inside the tools you already work in, such as Claude Code, Codex and Hermes, on the same shared allowance.

We will post here as each one ships. Questions and feedback are welcome at [email protected].