Space Bunny Alpha: Specs, Identity Clues and a Free API Guide
Space Bunny Alpha is a free, anonymous model on TokenRA with a 1,000,000-token context window and no public owner. It appeared on 23 September 2026 and nobody has claimed it. On this page we collect what is actually verifiable about it — the listing specs, the identity evidence published so far, and our own test results — and we label every claim with where it came from.
Last updated 24 September 2026 · See what changed
How we label sources on this page
Stealth models generate a lot of confident guessing. We separate what is stated by the provider from what has been independently measured from what is only a hypothesis.
What Is Space Bunny Alpha?
Space Bunny Alpha is a stealth release on TokenRA, listed under the model ID space-bunny-alpha. Stealth listings let a lab test a model in public under a codename before attaching its name to the result: the provider is hidden, the model is free, and the codename is the only identifier you get.
This one stands out for three reasons. It accepts video as well as image input, it carries a 1M-token context window, and it is free during the preview. There is no vendor page, no model card and no announcement — the TokenRA listing itself is the only first-party source of information, which is exactly why we mark everything else on this page with who said it.
Space Bunny Alpha Specs at a Glance
| Spec | Value | Source |
|---|---|---|
| Context window | 1,000,000 tokens | Provider-listed |
| Max output | 524,288 tokens | Provider-listed |
| Input types | Text, image, video | Provider-listed |
| Pricing | $0 — free preview | Provider-listed |
| Reasoning | Forced reasoning, five effort levels | Provider-listed |
| Tool calling | Supported | Provider-listed |
| Tokenizer | Classified only as "Other" | Provider-listed |
| Knowledge cutoff | Not published | Provider-listed |
| Provider | Not disclosed | Provider-listed |
Every row above comes from TokenRA's own listing. We have not independently verified the context window, the throughput or the pricing — no third party has yet published a reproduction of those figures.
Is Space Bunny Alpha MiniMax?
Not confirmed — and at the moment nobody can confirm it. The listing hides the provider, tags the tokenizer only as "Other", and leaves the knowledge cutoff field empty. There is no first-party field you can read a vendor name out of. That is the whole reason this question produces so much confident guessing: people are inferring an owner from behaviour, not reading it from a spec sheet.
What exists instead is indirect evidence, and the useful thing to do is to keep the evidence and the conclusion separate. The stealthprint project maintains a reusable fingerprinting toolkit for exactly this situation — it was originally written to analyse an earlier stealth model and has been pointed at this one. Its tokenizer probes point at the MiniMax tokenizer family. Independent That is the strongest public signal so far, and it is a measurement, not an announcement.
It is still not an answer. Matching a tokenizer family narrows the field; it does not name the lab, and it does not rule out a model that reuses a shared tokenizer. So the honest reading of the identity question today is:
- MiniMax M3.1? — Hypothesis the most-cited guess, supported by tokenizer-family evidence, unconfirmed by any vendor.
- What would settle it — a vendor claim, a fingerprint test that produces a distinguishing error pattern rather than a family match, or an independent evaluation with reproducible methodology.
- What would not settle it — another round of guessing on social platforms. The previous wave of stealth models produced a string of confident identity claims that turned out wrong; a short informal quiz that scored well in public collapsed when the test set got longer. Treat "it feels like a Chinese lab" as a prior, not a finding.
Benchmarks: What Has Actually Been Measured
This is where most pages on this model go wrong: they either publish the provider's marketing copy as if it were a result, or they repeat a benchmark without saying who ran it.
| Measurement | Who ran it | Status |
|---|---|---|
| Tokenizer fingerprint probes | stealthprint | Independent |
| Throughput and latency benchmarks | TokenDyno (Anton Gulin) | Independent |
| Context window, output limit, modality list, price | TokenRA listing | Provider-listed |
| Long-context retrieval, multimodal accuracy, date-knowledge probing | Various — see the model pages below | Mixed / unverified |
We deliberately do not quote a latency or tokens-per-second figure in this table. Benchmark numbers change as an endpoint is rerouted, and a number copied without its test conditions is worse than no number. Follow the two links above for live figures, and see our benchmark page for a per-test breakdown.
Our Own Tests
Everything above is somebody else's data, relabelled with its source. This section is where we put ours.
Status: first test published. Tested 24 September 2026 against the TokenRA OpenAI-compatible endpoint with model space-bunny-alpha. The API key is not stored or displayed here.
| Test | Result | Observed measurement |
|---|---|---|
| Short answer · non-stream | HTTP 200 · stop | 3,551.6 ms end-to-end · 300 completion tokens · 84.47 completion tok/s |
| Exact phrase · stream | HTTP 200 · stop | 1,435.7 ms to first SSE event · 1,448.4 ms total · 39 completion tokens · 26.93 completion tok/s |
Prompts: “In one short sentence, define a tokenizer fingerprint.” and “Reply with exactly: tokenizer fingerprint verified”. The non-stream response returned a visible 143-character answer plus a reasoning field; the stream response returned exactly tokenizer fingerprint verified.
Interpretation: Both endpoint availability and text generation passed. The completion-token rates are end-to-end observations calculated from the API's usage.completion_tokens divided by wall-clock duration; they include provider-side reasoning tokens where reported and are not a hardware-level generation benchmark.
If you have run a reproducible test on this model and want it included with attribution, tell us and we will link to your methodology.
The Stealth Alpha Pattern: Everything Gets Claimed
The most useful thing you can know about an unclaimed stealth model is how the previous ones ended. Every stealth Alpha release tracked since early 2026 has eventually been claimed by a named vendor, most of them Chinese labs, typically about a week after it appeared. The mappings widely reported so far:
| Codename | Later claimed as |
|---|---|
| Pony Alpha | Z.ai GLM-5 |
| Hunter Alpha / Healer Alpha | Xiaomi MiMo |
| Elephant Alpha | Ant Lingxi |
| Owl Alpha | Meituan LongCat-2.0 |
| Ox Alpha | Z.ai GLM-5.3-Flash |
| Union Alpha | Unbiased Pareto |
Codename-to-vendor mappings are recorded with evidence limits on our Stealth Alpha Models. Dates and source status are not established for every row.
That record is a base rate, not a prediction — it tells you the likely shape of the ending, not the name of the winner. It also tells you something practical: the free window is a countdown, not a feature. The previous stealth model in this series stayed free for roughly six days before it was revealed and repriced. If you are building on this model, design for the switch now rather than on the day it happens.
Free API and How to Use It
There is no separate signup, no model card to agree to and no payment method required while the preview lasts. You call it through TokenRA's OpenAI-compatible endpoint and set the model ID to space-bunny-alpha.
curl https://tokenra.io/v1/chat/completions \
-H "Authorization: Bearer $TOKENRA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "space-bunny-alpha",
"messages": [
{"role": "user", "content": "Explain what a tokenizer fingerprint is."}
]
}'
Three things worth knowing before you point anything real at it: the provider is anonymous, so do not send data you would not hand to an unidentified third party; the endpoint has already gone briefly offline at least once during the preview and come back, so treat availability as best-effort; and a free stealth listing can be revealed and repriced without notice. Step-by-step setup, including how to read the reasoning-level parameter, is in the API guide.
Space Bunny Alpha: FAQ
What is Space Bunny Alpha?
A free, anonymous model on TokenRA, listed as space-bunny-alpha. It supports a 1,000,000-token context window, accepts text, image and video input, and no company has publicly claimed it.
Is Space Bunny Alpha MiniMax?
Not confirmed. TokenRA lists the tokenizer only as "Other" and publishes no knowledge cutoff. The strongest public signal is third-party tokenizer fingerprint testing, which points to the MiniMax tokenizer family — but no vendor has claimed the model. Treat any identity claim as a probability, not a fact.
Is Space Bunny Alpha free?
Yes — the listing shows $0 as a free preview at the time of writing. Free stealth previews have historically ended once the model was claimed.
What is its context window?
1,000,000 tokens, with a maximum output length of 524,288 tokens, as stated by the provider.
How do I use it through an API?
Get a TokenRA API key, point your client at the OpenAI-compatible endpoint and set the model ID to space-bunny-alpha. No separate signup or payment method is required while the free preview lasts.
Will it be claimed by a company?
Historically, yes — every stealth Alpha model tracked since early 2026 has been claimed, most by Chinese labs, typically about a week after appearing. That base rate is not a prediction for this specific model.
Latest Updates
- 24 Sep 2026 — Page published: listing specs, source labels and the evidence round-up.
- 23 Sep 2026 — Users report the model briefly went offline on TokenRA and came back after the issue was resolved. Third-party report
- 23 Sep 2026 — Space Bunny Alpha appears as a stealth listing with a 1M-token context window and video input. Provider-listed
This section is updated whenever something verifiable changes. If a vendor claims the model, it gets its own page and this page becomes the record of the codename.
All Space Bunny Alpha Pages
- Benchmarks
- Is It MiniMax?
- API Guide
- What Is It?
- How to Use It
- Free Tier and Limits
- vs MiniMax M3
- Context Window - coming soon
- Tokenizer Fingerprint - coming soon
- SVG Generation Test - coming soon
- Review - coming soon
- vs Claude - coming soon
- Stealth Alpha Models