Reasoning models think in the dark.
Most bills keep it there.
Ours don't.
A one-line question can burn thousands of thinking tokens before the model answers. Wallaby is the curated inference layer for frontier open-weight models — Kimi K3 live today — that prints the full calculation on every call.
-H "Authorization: Bearer $WALLABY_API_KEY" \
-d '{"model":"kimi-k3",
"messages":[{"role":"user","content":"gday"}]}'
# using the OpenAI SDK?
# change base_url. that's the whole migration.
Where a token bill goes dark
THREE WAYS IT HIDESThe lump sum.
Thinking, cache and output folded into one number. You pay it because you can't check it.
The rerun tax.
A weaker model answers wrong; you prompt again. Cheap per token, expensive per accepted answer.
The vanishing endpoint.
Grey-market keys, no company behind them. No invoice, no recourse, gone next quarter.
The price list
USD PER 1M TOKENS · OFFICIAL LIST VERIFIED MONTHLY| Model | Wallaby (flat) | Official list | You save |
|---|
Thinking tokens — the model's reasoning before it answers — bill as output, same as the official API, and every receipt shows them on their own line.
Questions on pricing, volume or billing? Talk to us — we reply within 48 hours. Over $500 a month: volume pricing via support@wallabytoken.com.
Is this the real Kimi K3? Yes — open weights under the Kimi K3 License, verified against the official API.
Do you log my prompts? No — content logging is disabled at the gateway; we store only what billing requires.
What happens if you go down? The gateway fails over to a backup provider automatically.
Can I get a refund? Prepaid balance is generally non-refundable; billing errors and your Australian Consumer Law rights are the exceptions.
What you can count on
FOUR PROMISES, KEPTEvery call, itemised.
Input, cached input, output, thinking — four lines on every receipt: dashboard, API response, invoice. No content logs, no training on your data; we keep only what billing requires.
Open weights, on purpose.
Frontier open-weight models we'd run ourselves — Kimi K3 today, more when they earn the spot. Inspect, fine-tune, self-host one day: the model you build on can't be repriced or withdrawn overnight. One model served honestly beats twenty served vaguely.
Failover built in.
Traffic routes across upstream providers and fails over automatically when one goes down. You don't notice. Your users don't notice.
A real company behind the key.
Wallaby Data Pty Ltd — registered in Australia, billed in USD via Stripe, itemised invoice with every top-up, refunds per Australian Consumer Law. No crypto, no grey-market accounts.
One prompt. One snake game. Thirteen cents.
EVERY TOKEN, ITEMISEDReasoning models think before they answer, and thinking costs money. The receipt on the right is real — one prompt, 4 Sep 2026, one playable snake game. 131 tokens in, 9,776 out, 6,572 of them spent thinking. Two-thirds of the bill, on its own line.
The same prompt ran again and barely thought: $0.040. A bonus, not a promise.
Same model. Same task. Different thinking — that's why we show the bill.
Play the game this prompt built →
A real prompt, made 4 Sep 2026. Every run thinks a little differently — treat it as a reference, not a quote.
Same prompt, second run. Play this one too →
Sixty seconds
REGISTER → $0.50 → API CALLRegister
Email and password. No phone, no card until you top up.
Get your key
$0.50 free credit is already on your account — enough for dozens of test calls.
Point your code at us
Swap base_url to https://api.wallabytoken.com/v1 and use your key. If it speaks OpenAI, it speaks Wallaby — key, balance and receipts stay put.
Full reference — streaming, tool use, structured outputs, parameters, error codes: API Docs.
Top up
PREPAID · NEVER EXPIRESWhat $20 buys: about 150 calls like the snake-game receipt above — $0.1324 for a playable game, itemised to the token.
Card payments are live via Stripe — pick an amount to check out from the console. Self-serve top-ups run $20 to $500; larger volumes and invoicing: support@wallabytoken.com.
Dispatch
FIELD NOTESPricing comparisons, agent setups, memory systems and post-mortems from running this gateway in production — we write down what we run. Read the blog →
Kimi K3 API pricing: $3/$15 vs 17 providers
Every route we priced, side by side — what a million tokens actually costs, and where we land.
SetupKimi K3 in Cline: 5-minute BYOK setup
Bring your key, point Cline at the endpoint, and be shipping code five minutes later.
TroubleshootingCline + Kimi K3: every error we've hit, and the fix
The error map from production — auth failures, rate limits and bad model names, each with its fix.