Frontier AI models in one place.
One API key.
Kova puts Kling, Hailuo, Seedance, DeepSeek, Qwen, GLM and Kimi — behind a single OpenAI-compatible endpoint, billed in USD, at a target of roughly a third of Western list prices. No Chinese phone number. No real-name verification. No RMB payment rails.
BASE URL AT LAUNCH https://api.kovamodel.com/v1 · swap the base URL, keep your SDK
▶
▶
▶
The showcase
Every model will be reachable through the same key. Prices are the planned launch list in USD — the target rates for opening day, confirmed with each upstream before the first key is issued. No seat fees, no minimums.
The models are cheap.
Getting to them isn't.
Chinese labs price for their home market, in RMB, behind signup walls built for Chinese residents. That gap — not the technology — is what you're paying Western prices for.
Direct signup, if you try it yourself
Each provider is its own account, its own wallet, its own rules:
Through the gateway
We operate the domestic accounts, the verification and the routing. You just call an API:
- One key, every provider — Kling to Kimi through a single OpenAI-compatible endpoint.
- Priced off Western alternatives, not off cost — comparable output at roughly a third of Veo, Sora or Claude list prices, with the maths shown in the open below.
- Multi-upstream routing — where a model exists on more than one upstream, traffic shifts automatically when one throttles or goes down.
- Prepaid and capped — top up $10–$50 at a time. No large balance ever sits in the system, yours or ours.
- Built to be tested in public — hands-on tests of these models run on the Kova channels, so the catalog and the content never drift apart. Real prices, real tests.
Same job. A fraction of the bill.
Planned launch prices against the Western model you'd otherwise use for the same task.
| Task | Western route | Via Kova | You save |
|---|---|---|---|
| 5-second 1080p video cliptext-to-video, with audio | Veo 3 — $3.755 s × $0.75 / sec | Kling 2.5 Turbo — $0.35 | −91% |
| 1M output tokens, frontier reasoningchain-of-thought, tool use | Claude Sonnet — $15.00per 1M output tokens | DeepSeek R1 — $2.30 | −85% |
| Agentic coding, 1M input tokenslong-context repo work | Claude Sonnet — $3.00per 1M input tokens | Kimi K2 — $0.65 | −78% |
| One 4K imagetext-to-image, commercial use | GPT Image — $0.17high quality, per image | Seedream 4.0 — $0.035 | −79% |
Western figures are public list prices for the named models; equivalents vary by resolution, duration and options. Gateway figures are the planned launch list — see the catalog for the unit behind each price.
Price a call before you make it
Pick a model, write a prompt, see exactly what the call would cost. The demo runs on a $10.00 sample balance — every run deducts the catalog target price.
will appear here after a run.
Live in four lines of code
The gateway speaks the OpenAI API dialect. If your app already calls OpenAI, Anthropic-style or a local SDK, integration is a base-URL swap and a model id.
Join, then get your key
Keys go to founding members first, issued through the community while the public dashboard rolls out. Once yours lands, top up $10–$50 — one key works for every model in the catalog.
Point your SDK at the gateway
Set base_url to the URL above. Your existing OpenAI SDK, LangChain or LiteLLM setup keeps working unchanged.
Swap the model id
Use catalog ids like kling-2.5-turbo, deepseek-r1 or kimi-k2. Video and image calls are async: submit, poll, download.
Watch one bill
Every call — any provider — lands on a single USD usage ledger, priced at the catalog rate shown before you ran it.
Prepaid. Capped. No surprises.
Usage will be billed at the catalog target rates. Plans only change your rate lock, queue priority and support — never what's available.
For testing the water. Top up a card, call anything.
- All 13 launch models
- Top-ups of $10–$50, card in USD
- Catalog launch rates
- Community support (Skool)
The founding cohort rate — locked for life while the plan stays active.
- Launch rates locked — no repricing on you
- Priority queue on video renders
- Higher rate limits & concurrency
- New models in beta before public listing
- Direct support from the operator
For teams pushing real volume through video or agents.
- Volume pricing below catalog
- Multiple keys & per-key budgets
- Dedicated routing capacity
- Monthly invoicing
Why capped top-ups? A gateway should never be a place where a big balance can get stuck. Small, frequent top-ups keep your exposure — and ours — close to zero. It's a design choice, not a limitation.
Before you ask
Do I need a Chinese phone number or ID?
No. That's the point of the gateway. Accounts, real-name verification and RMB settlement happen on our side, under business accounts operated for this service. You sign up with an email and a card, like any Western API.
Is it really OpenAI-compatible?
Yes for chat and language models: same request shape, same streaming, same tool-calling format — change the base URL and the model id. Video and image models use a simple async submit-and-poll flow on the same key and base URL, documented in the Docs section above.
Why are the prices so much lower than Veo or Claude?
Two reasons. Chinese labs price aggressively for their domestic market, in RMB — DeepSeek's API famously costs a fraction of US equivalents. And several of the best models (Kling, Hailuo, Seedance) barely compete for Western API customers at all, so there's no Western price umbrella over them. We anchor our prices to what the Western alternative costs, not to our upstream cost — you get roughly two-thirds off, we keep a sustainable margin.
What happens if an upstream provider goes down?
Where a model is served by more than one upstream, the gateway reroutes automatically. If a model is fully down, calls fail fast with a clear error and are not billed — you're never charged for a render or completion you didn't receive.
How do payments and refunds work?
Prepaid credits in USD, bought by card, in $10–$50 top-ups. Credits are spent per call at catalog rates. Unused credit on a closed account is refundable to the original card — the capped top-up design means that amount is never large.
Who runs this?
Kova — the team behind the channels that publish real price breakdowns and hands-on tests of these models across YouTube, X, Instagram, Reddit and dev.to under @kovamodels. The gateway is the same work, productised: we were already testing every one of these models; now you can call them.
One key. Every frontier model.
A third of the price.
Founding members lock launch rates for life and get new models first. Keys are issued through the community while the public dashboard finishes rolling out.