XavaInferencesign indocs
one key · every model · discounted

Every AI model. Discounted.

Pay less than going direct to OpenAI, Anthropic, Google and 40+ others. One key for text, image, video and audio. Allocate your discount as savings or token buybacks.

get a keybrowse 229 modelsread the docs
discountedevery model, cheaper than going direct
229models behind one API key
5%off when you pay in USDC
0prompts stored, ever
A generated poster in photocopy style: huge stencil letters XINF beside a server rack whose cables feed a chain of blocks; the last block is highlighter green with a check mark, and a red PROOF stamp sits in the cornerimage
OpenAI GPT Image 21536 x 1024 poster, unedited$0.1328
video
ByteDance Seedance 2.530 s explainer with dialogue, 720p$5.55

“This voice came through Xava Inference. Every AI model, one key, discounted. Allocate your discount as savings or token buybacks.”

audio
ElevenLabs v3text to speech, 130 characters$0.0104

promptWhat is Xava Inference? One sentence, under 25 words.

Claude Opus 5Xava Inference sells prepaid AI credits for 200+ models through one OpenAI-compatible API key, cheaper than going direct, with your discount as savings or token buybacks.

text
Claude Opus 5406 tokens$0.0026
how it works

Ready for humans and agents.

Three steps either way. Do both from one account: your keys, your agent's calls and your split all live together.

get access
01you

Sign in with Google or a Solana wallet and create an API key.

your agent

Install our plugin with one copy and paste, or skip the account and pay per call in USDC with x402.

Top up by card, or with USDC on Solana and save 5%. From $5.00.

choose your split
02you

Allocate your discount as savings or token buybacks: a lower price, buybacks of any Solana token you pick, or a mix.

your agent

Calls made with your key follow your account split. x402 calls can set a split per request, and keep the full discount if they don't.

One setting for your whole account. Change it any time.

call any model
03you

Point any OpenAI-compatible SDK at us. Text, image, video, audio and embeddings.

your agent

It picks the right model, tells you the price first and stays under the budget cap you set.

Every buyback your usage funds shows up on the public proof page.

pricing

Standard price in. Your price out.

Every model starts at its standard price, what the model maker charges when you buy direct, minus your discount. No markup, no platform fee on top.

example
$1.00→$0.8020% off

$1.00 of usage at the standard price costs you $0.80 with the whole discount taken as a lower price, which is the default.

Example: a model with 20% off. Each model has its own discount. Or send part of your discount to buybacks. One setting for your whole account, applied to every model; change it any time.

split your discountpreview
0%
pick a token
0%
100%

On $1.00 at standard price: you pay $0.80 · $0.00 buys $XINF · $0.00 buys TOKEN

$XINF × $XAVA

Holders earn credits. Stakers earn USDC.

Every credit is backed by USDC. When credits are spent, our margin splits 50% into $XINF buybacks for a lock that is never sold and 50% to $XAVA stakers in USDC. No team share. The lock's own earnings become AI credits for $XINF holders.

$XINF holders
3%

of every $XINF transfer is paid to holders, pro rata, in tokenized NVIDIA stock (NVDA).

Paid automatically every few minutes. Holders also earn AI credits from what the lock earns, claimed weekly: earn AI credits.

the lock
33%

of $XINF supply is bought at launch and locked for good. 50% of our margin keeps buying $XINF into the same lock.

Locked $XINF is never sold. Its tokenized NVDA is swapped to USDC into the credit reserve and becomes AI credits for holders.

$XAVA stakers
50%

of our margin, in USDC, to $XAVA stakers on Avalanche by time-weighted stake, up to 3x for stake held 180 days.

Claim weekly on Rewards within 60 days of each list. Unclaimed $XINF holder credits are burned through the same split.

Proof of reserves and every burn batch: reserves.

live

Public numbers, refreshed every minute.

These fill in when the API opens. Every figure comes from the same ledger that bills you.

models229
requests 30d1
tokens 30d20
saved by users$0.000000212
$XINF bought$0.00
other tokens bought$0.00
$XINF locked—
USDC to $XAVA stakers—
credit reserve—
faq

Questions, answered briefly.

What am I buying?
Prepaid credit for AI inference, spent per request at each model's standard price (what the model maker charges when you buy direct) minus your discount. No subscription, no seat fees. Minimum top-up $5.00.
Where does the discount come from?
We source inference capacity for less than the standard price and pass most of that saving to you. Each model has its own discount, shown as % off next to its standard price in the catalog.
What does the split do?
Allocate your discount as savings or token buybacks: it can come off your price, or buy back a Solana token you choose, in any mix. It is one setting on your account, applied to every request. Buybacks are batched and executed on-chain, and every batch is on the public proof page.
What exactly is a credit?
A credit is $1 of prepaid inference, held as a non-transferable token on Solana. Every credit is backed 1:1 by USDC in a public reserve: credits are minted only when USDC arrives in the reserve (your top-up, or the lock's earnings swapped to USDC), and the reserve is always at least the credit supply. Usage is metered live and settled in hourly burn batches. Reserves and every burn batch are public on the reserves page.
Where does the platform margin go?
When credits are spent, the USDC behind them pays the inference cost, then any buybacks you chose in your split. What is left is our margin, split at every burn: 50% buys $XINF into the permanent lock and 50% goes to $XAVA stakers in USDC. There is no team share. If an affiliate referred you, their cut comes out of the $XINF half: 10% of margin if they take USDC, 20% if they take credits.
What do $XINF holders get?
Two things. First, 3% of every $XINF transfer is collected and paid to holders pro rata in tokenized NVIDIA stock (NVDA), automatically every few minutes: nothing to stake or claim, just hold. Trading pools are not paid; their share goes to holders. Second, the protocol lock earns that same 3% on its $XINF; its NVDA is swapped to USDC into the credit reserve and becomes AI credits for $XINF holders, shared by time-weighted balance every hour and claimed once a week on Rewards. To earn credits in an hour, your average balance over that hour must be at least 1,337 $XINF; eligible wallets share the pool pro rata by that time-weighted balance, and the share of wallets below the floor goes to the eligible ones. Credits are for using models and cannot be cashed out.
What is the $XINF lock?
33% of the $XINF supply, bought at launch, plus every $XINF buyback funded by our margin. It is locked on-chain for good and never sold: the lock program has no withdraw instruction. It earns the 3% holder rewards in tokenized NVDA like any other holder, and what it earns becomes AI credits for $XINF holders.
What do $XAVA stakers get?
50% of our margin, in USDC, every time credits are burned. Your share is your credited stake in the Avalaunch staking contract on Avalanche C-chain, weighted by time and by a loyalty multiplier: each deposit or compound starts its own tranche at 1x, rising steadily to 3x over 180 days. A partial withdrawal removes your newest tranches first, so older tokens keep their multiplier; a full withdrawal resets everything. Only staked $XAVA on Avalanche counts; holding without staking does not.
How do $XAVA stakers claim?
On Rewards: connect the Avalanche wallet you stake from and link a Solana wallet. You sign one plain message with both wallets saying where your rewards go. Each week's amount is then claimed on-chain in USDC. Every staker counts by the same rule. Changing your payout wallet takes 48 hours to apply. On-chain claims open at launch; the weekly lists are published before that.
How do I know the reward numbers are right?
Every week we publish what each wallet can claim, with a fingerprint anyone can recompute, and the daily stake weights behind it. Every burn batch is published with its hash and the usage log it settles. Funding waits several hours after publishing, so an independent check can stop a wrong list.
What if someone never claims?
Each week's list can be claimed for 60 days. Unclaimed $XINF holder credits expire after 60 days and are burned: the full USDC behind them goes through the same split, 50% buys $XINF for the lock and 50% goes to $XAVA stakers. Unclaimed $XAVA staker USDC is added to a later staker pool. Nothing goes back to us. See Rewards.
What are this week's deals?
Each week we may announce extra discounts on specific models for $XAVA stakers, on top of the usual discount, paid for out of our margin and never larger than it. You qualify when your account has a linked Avalaunch staking wallet with stake in it; a new stake counts within about 10 to 15 minutes. Current deals are marked "this week" in the model catalog.
Are the programs audited?
The lock and credits programs are our own small open-source Solana programs. They are independently audited before they go to mainnet, and nothing goes live on mainnet without that. See the whitepaper.
Can the lock's rewards change or stop?
Yes. The launchpad's contracts are not immutable: it sets the 3% rate and decides whether the lock is included in holder payouts, and it could change or stop either. Tokenized NVDA's issuer can also freeze, pause or take back tokens. See the risks in the whitepaper.
Am I getting the real model?
Yes. Your request goes to the model you name. Nothing is swapped, downgraded or quietly rerouted to a cheaper model.
Do you store my prompts?
No. We keep billing metadata only: which model, token or unit counts, cost and timing. Prompts and responses are never stored by us.
What code changes do I need?
Two lines: the base URL and the key. Any OpenAI-compatible SDK works, and so does plain HTTP. Image, video and audio models use the same key.
How do I pay?
By card at our posted price, or with USDC on Solana and save 5%: $100 of credit costs $105 by card or $100 in USDC. Agents can skip the account entirely and pay per request with x402.
Can I get a refund?
Credit is prepaid and not redeemable for cash. See the terms for the details.
Can I sell capacity?
Soon. Sellers will list inference capacity at a discount, and requests route to the best price with automatic failover.
start

Get a key. Keep the change.

229 models, each cheaper than going direct. Allocate your discount as savings or token buybacks.

get a keybrowse modelswhitepaper