Quick disclosure: we run Wallaby Token, an AI inference platform providing metered access to open-weight models. This post explains why the company exists. You deserve to know our incentive before our pitch.
The problem we kept hitting
Frontier open-weight models are good enough to build real products on. Buying access to them is still harder than it should be. If you're a developer who wants Kimi K3 in your coding tool today, your options are a provider account whose billing may not take your card — or a third-party marketplace that marks up tokens and shows you a balance going down with no explanation.
We wanted a third option: one prepaid USD key, an OpenAI-compatible endpoint, and a bill you can actually read. It didn't exist, so we built it.
What we run
One key. Prepaid USD. Per-token metering. If your tool speaks to api.openai.com, it speaks to us — change the base URL and keep working:
from openai import OpenAI
client = OpenAI(
base_url="https://api.wallabytoken.com/v1",
api_key="sk-...",
)
Pricing: Kimi K3, live now
| Input | Cache hit | Output | ||
|---|---|---|---|---|
| Official list | $3.00 | $0.30 | $15.00 | per 1M tokens |
| Wallaby | $2.70 | $0.27 | $13.50 | 10% off list, ongoing |
No purchase fee, no minimum beyond a $20 first top-up. The price on the page is the price you pay.
How we're different from a typical API platform
- The receipt is the product. Every call is itemised: tokens in, tokens out, unit price, total. When a coding agent burns a long session, you can answer "what did that cost me?" in seconds.
- Open-weight models only. Not because closed models aren't good, but because open weights are where your investment can't be revoked by a terms-of-service change — and where a focused provider adds real value. We wrote about why provenance became a procurement question: the Authors Guild filings are an API procurement problem.
- We don't chase model hype. The catalog stays short: frontier open-weight models we've actually run in production ourselves. No landing pages for models we haven't tested.
Where we're going
- More frontier open models, fast. When a model worth carrying drops, the goal is live-in-48-hours with real numbers — not a waitlist.
- Field notes. This blog publishes what we measure: cross-provider Kimi K3 pricing, smoke-test results, honest caveats. Numbers over adjectives.
- Boring reliability. Uptime, auditable metering, support that answers. The unglamorous work that makes an API provider worth paying.
We're early, and some of this is still being built — we'd rather say so than fake it.
Try it: grab a key. New accounts get $0.50 of free credit, and your first call shows up as an itemised receipt within seconds. That's the whole pitch.
— The Wallaby team, Sydney