Hop after hop
Your request is relayed from layer to layer before a model sees it. Every hop adds time before the first token — and one more place to break.
One gateway for the AI models you build on
TokenFusion is a single OpenAI-compatible gateway to leading AI models. One API key, one prepaid balance in US dollars, and every call streamed straight through — without the marketplace detour that costs you speed, capacity and clarity.
The aggregator detour
Many teams reach models through marketplaces and routers that sit on top of other routers. Each layer is one more wait before your first token, one more shared pool to queue in, and one more place a call can fail.
Your request is relayed from layer to layer before a model sees it. Every hop adds time before the first token — and one more place to break.
Shared keys and shared pools turn someone else's traffic spike into your timeout. The throughput you tested on Monday is not there on Tuesday.
Prepaid balances that expire, rounding you can't follow, and no record of what each call actually cost you.
How TokenFusion works
Your key is verified and the call's worst case is held against your balance. Calls are held against your available balance, and a call needs at least $1.00 available to start — nothing ever runs into debt.
We pass your request straight to the model and stream every token back the moment it's generated. Nothing waits in a queue in the middle.
When the call ends you're charged at the published selling rate for the tokens it reported, and the rest of the hold is released straight back to your balance.
Built for concurrency
In a shared pool, one customer's spike becomes everyone's error. TokenFusion gives every account its own fair share of calls in flight, so no single account can take all the capacity — and a burst waits at its own gate instead of knocking everyone else over.
Each account has its own limit on calls in flight, checked before anything is held or charged.
At a limit you get an immediate 429 or 503 with
Retry-After, so your client knows exactly when to try again.
Capacity is always kept back for sign-in, top-ups and your dashboard, even when calls are at their peak.
What you get
Point your existing OpenAI SDK at one base URL. Chat, streaming and image models sit behind the same endpoint.
Create a key for any model in a click, with ready-to-paste code. Revoke one without touching the rest.
Top up from $1. Your balance never expires, and it cannot go negative.
Every call is charged at the published selling rate for its model, shown to six decimal places, with per-model totals.
Fund each team from your wallet, cap each member by the month, and see who used what.
Every model states the region it is served from. If we have not been told, we say so — we never guess.
If a model cannot be served, the call is refused and nothing is billed. You never get a made-up answer.
A receipt for every top-up and a per-call usage history you can check line by line.
Pricing
Pick what you're doing and how often. Every figure comes from our live rate card, in US dollars, worked out exactly the way a call is billed. You are charged at the published selling rate shown for the model, on the usage reported for each call. If a call comes back with no usage, it is charged on an estimate — and marked as one on your usage page.
Pay $500 and $540.00 lands in your balance, so every call on every model costs you less.
Your balance never expires. Nothing you top up is wasted on a deadline.
No subscription, no seats, no minimum term. Each call is charged on the tokens it reports, to the millionth of a dollar.
If a model cannot be served, the call is refused and nothing is billed.
| Model | Input /1M | Output /1M | Cost per call | Per month |
|---|---|---|---|---|
| Gemini 1.5 Flash · global-xh-05Not available yet
Google · Premium · 1M context |
$0.075 | $0.30 |
$0.000165 |
$1.65 |
| GPT-4o mini · global-xh-02Not available yet
OpenAI · Premium · 128k context |
$0.15 | $0.60 |
$0.00033 |
$3.30 |
| Llama 3.1 70B · global-xh-09Not available yet
Meta · Premium · 128k context |
$0.35 | $0.40 |
$0.00047 |
$4.70 |
| DeepSeek Chat · global-xh-08Not available yet
DeepSeek · Premium · 64k context |
$0.28 | $1.10 |
$0.00061 |
$6.10 |
| GLM-4 Plus · global-xh-07Not available yet
Zhipu · Premium · 128k context |
$0.60 | $0.60 |
$0.00078 |
$7.80 |
| Gemini 1.5 Pro · global-xh-04Not available yet
Google · Premium · 1M context |
$1.25 | $5.00 |
$0.00275 |
$27.50 |
| Qwen Max · global-xh-06Not available yet
Alibaba · Premium · 32k context |
$1.60 | $6.40 |
$0.00352 |
$35.20 |
| GPT-4o · global-xh-01Not available yet
OpenAI · Premium · 128k context |
$2.50 | $10.00 |
$0.0055 |
$55.00 |
| Claude 3.5 Sonnet · global-xh-03Not available yet
Anthropic · Premium · 200k context |
$3.00 | $15.00 |
$0.0075 |
$75.00 |
Each figure is what one call with exactly these token counts is charged on that model; your real charge follows the tokens each call reports. Per month is one call times the number of calls. A crossed-out figure is the same call at the model's list price, shown only where your price is lower.
Any whole-dollar amount from $1 to $10,000; bigger top-ups add a bonus to your balance. No contracts, no minimum term. Your balance stays on the account for as long as the account is open.
See the full rate cardOpen an account in a minute, top up, and call the model you need with one key.