Space Bunny Alpha: What It Is, Why It Looks Like MiniMax M3.1, and How to Use It
Track Space Bunny Alpha's anonymous OpenRouter and OpenCode release, reported 1M context and 524K output, community feedback, tokenizer evidence, MiniMax M3.1 Flash theory, and API alternatives.
Published 2026-09-28
Space Bunny Alpha is an anonymous free-preview model, not an officially named MiniMax API. It appeared around September 23, 2026 as stealth/space-bunny-alpha on OpenRouter and space-bunny-free in OpenCode. Community reports describe roughly 1M context, up to about 524K output, text/image/video input, text output, and adjustable reasoning effort from low through max. OpenRouter and OpenCode usage surged after launch, but neither MiniMax nor another model vendor has officially claimed the model.
The strongest current attribution is MiniMax-family, with M3.1-Flash-Preview as the leading version theory. Independent tokenizer probes reportedly match MiniMax vocabulary at 24/24 and 50/50 while matching GLM much less often. The 1M context and reasoning controls also resemble the M3.1-Flash-Preview that MiniMax opened in MiniMax Code on September 27. This is strong family evidence and a plausible version hypothesis, not a MiniMax confirmation. Tokenizer tests cannot distinguish M3, M3.1, a special serving build, or a gateway template.
Space Bunny has no confirmed public APIMaster route or official price. For a callable MiniMax API today, APIMaster's live minimax-m3 card has 3 active routes, starting at $0.30 input / $1.20 output per 1M tokens against $0.60 / $2.40, or about 50% off when checked September 28, 2026. The live card is the source of truth.
| Model ID | Original price (1M input / output) | APIMaster current price | Active routes | Discount | Model card |
|---|---|---|---|---|---|
space-bunny-alpha |
Not published | No public route | 0 | — | View model card |
| M3.1-Flash-Preview (MiniMax Code label; API ID unannounced) | Not published | No public route | 0 | — | View model card |
minimax-m3 |
$0.60 / $2.40 | From $0.30 / $1.20 | 3 | About 50% off | View discount |
GPT Trial(GPT-6 Astra, GPT-6 Sol, GPT-5.6, and GPT-5.5 available)
Sign up and claim a $20 GPT trial
What is Space Bunny Alpha?
Space Bunny Alpha is the community name for a free, stealth-routed model that appeared on OpenRouter and OpenCode in late September 2026. The provider-facing labels are stealth/space-bunny-alpha on OpenRouter and space-bunny-free on OpenCode. It is best understood as a preview endpoint or anonymous serving identity, not as a stable public model family.
The model drew attention because it combined unusually large reported limits with low latency and free access. It was used for coding agents, repository work, small applications, SVG and UI generation, and multi-step product changes. A high request volume does not prove model quality or lineage, but it does explain why the name spread quickly through developer communities.
Space Bunny specs and access
The following figures are community observations and provider-facing behavior, not an official model card:
| Reported property | Current community description | Confidence |
|---|---|---|
| Context window | About 1M tokens | High as an observed route property; not vendor-published |
| Maximum output | About 524K tokens | Reported; verify against the current gateway limit |
| Inputs and output | Text, images and video in; text out | Reported route behavior |
| Reasoning | Commonly enabled; effort from low to max |
Reported interface behavior |
| Price | Free during preview on the named gateways | Gateway policy, not a permanent API promise |
| Official model card | None from MiniMax or another vendor | Not published |
The reasoning setting matters in practice. A high effort setting can produce better multi-step coding work, but it can also spend too long on a greeting or a small JSON response. Treat the effort parameter as a latency and reliability control, not a universal quality slider.
Why people think it is MiniMax
The attribution has several independent-looking pieces, but they do not all prove the same thing.
| Evidence | What it supports | Limitation |
|---|---|---|
| Tokenizer probes reported at 24/24 and 50/50 for MiniMax | The vocabulary looks like the MiniMax family | It cannot identify M3.1 versus another MiniMax build |
| GLM probes reportedly matched only 7/24 | Weakens the GLM-5.3-Flash theory | A probe result is not a vendor attestation |
| Kimi and M2.5 comparisons | Makes some alternative family theories less likely; M2.5's reported context is much smaller | Negative tests depend on the probe design and gateway behavior |
1M context, multimodal input and low–max effort |
Resembles the reported M3.1-Flash-Preview shape | Similar product controls can be copied by other serving stacks |
| Timing after the September 27 MiniMax Code preview | Fits a stealth preview or stress-test channel theory | Timing alone does not establish identity |
The right conclusion is high confidence that Space Bunny is MiniMax-family, medium confidence that it is an M3.1-Flash-Preview anonymous channel. It is not accurate to call it confirmed M3.1, and the current evidence does not support the GLM attribution.
Space Bunny versus MiniMax M3.1 Flash Preview
MiniMax opened M3.1-Flash-Preview in MiniMax Code on September 27 for everyday development. That product label and Space Bunny's anonymous endpoint have overlapping reported traits: a large context window, multimodal input, adjustable effort and a focus on coding agents.
| Question | Space Bunny Alpha | M3.1-Flash-Preview |
|---|---|---|
| Where it appeared | OpenRouter and OpenCode anonymous labels | MiniMax Code product UI |
| Public API ID | Not confirmed | Not announced as a stable public API ID |
| Vendor attribution | Not claimed | MiniMax product label |
| Context and output | About 1M / about 524K reported | Public full limits not documented in the announcement |
| Price | Free preview on named gateways | MiniMax Code plan access and campaign terms |
| Relationship | Leading theory: an anonymous MiniMax preview channel | Confirmed MiniMax Code preview; exact connection unconfirmed |
The two names may refer to the same underlying model, a specialized serving build, or related checkpoints. Until MiniMax publishes a model card or confirms the endpoint, keep the names separate in production documentation.
Community feedback: why developers like it
The positive reports focus on a practical combination: speed, low cost and enough competence to keep an agent moving. Public route tests commonly describe roughly 70–120 tokens per second with low time to first token, although latency varies by gateway, queue and effort setting.
Developers have reported useful results on small tools, bug fixes, repository changes and real product iterations. Community examples include a Mac full-screen alarm, drag-and-drop fixes, an on-chain radar built from live data and screenshots, and SVG or interactive UI work. Some users say the model stays coherent across long repository tasks and cross-file refactors instead of losing the original constraints.
The usage spike also has a non-technical explanation: it was free during preview, and OpenRouter prominently described a no-training-on-prompts policy for that route. Usage figures reported for September 27 reached several trillion tokens across the two gateways. Those numbers show demand, not a benchmark win.
Complaints and limits
Space Bunny is not a reliable “do exactly six lines” model. The most repeated complaints are:
- Overthinking: high effort can make trivial prompts take minutes and can cause external parsers to time out.
- Format obedience: one independent suite reportedly scored 128/157; coding cases included syntax errors and strict-length failures such as returning 129 lines for a six-line request.
- Uneven simple tasks: some OpenCode users reported failed HTML/CSS/JavaScript mini-games even when other models completed them.
- Operational volatility: the free route experienced a short outage around launch, and gateway retention policies differ.
- Mid-pack aggregate benchmarks: public leaderboard placement is not consistently frontier-level; the appeal is free speed and agent fit.
Use Space Bunny for exploratory coding and agent work where retries are cheap. Do not make it the only route for strict JSON, exact-length generation, safety-sensitive actions or a production workflow that cannot tolerate anonymous-provider changes.
Space Bunny effort and agent workflow
For an agent, effort selection should be explicit:
| Workload | Starting effort | Why |
|---|---|---|
| Greeting, classification or small JSON | low |
Avoid unnecessary reasoning and timeout risk |
| Small bug fix with tests | medium |
Balance repair quality and latency |
| Multi-file feature or repository audit | high |
Allow more planning and cross-file checking |
| Difficult autonomous coding task | xhigh or max |
Reserve for cases where extra latency is acceptable |
Add a timeout, retry and output-size guard around the model. Validate JSON before parsing, keep tool calls idempotent, and make the model configurable so a free preview can be replaced without changing your worker. For side effects such as commits, purchases or production deploys, require an explicit approval step.
Can you call Space Bunny through an API?
There is no confirmed public Space Bunny API model card or APIMaster route at the time of this check. The OpenRouter and OpenCode labels are gateway-specific identities, and a free preview can change its limits, retention terms or availability without the stability guarantees of a documented vendor API.
If you need a callable MiniMax API today, use the live minimax-m3 route on APIMaster. It is a different model and should not be advertised as Space Bunny or M3.1. Register, create a key, inspect the live card and run a small fingerprint and workload test before sending production traffic:
curl "https://apimaster.ai/v1/chat/completions" \
-H "Authorization: Bearer $APIMASTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"minimax-m3","messages":[{"role":"user","content":"Inspect this failing test and propose a minimal patch."}]}'
The current minimax-m3 price is a marketplace snapshot, not a Space Bunny price. Check the live M3 discount card immediately before use.
What developers should verify next
Before treating Space Bunny as M3.1, look for four kinds of evidence:
- MiniMax explicitly names Space Bunny or the gateway model ID in an official announcement.
- A public model card states context, output, modalities, effort controls, retention and training policy.
- A stable API identifier and price table appear in vendor or gateway documentation.
- Independent fingerprint tests reproduce the result across more than one route and over time.
Until then, describe it as MiniMax-family stealth preview, most likely related to M3.1-Flash-Preview. That wording preserves the useful signal without turning a community attribution into a fact.
FAQ
What is Space Bunny Alpha?
Space Bunny Alpha is an anonymous free-preview model listed as stealth/space-bunny-alpha on OpenRouter and space-bunny-free in OpenCode. It is not an officially documented model from MiniMax or another vendor.
Is Space Bunny MiniMax M3.1?
Probably related, but not confirmed. Tokenizer probes strongly point to the MiniMax family, and the reported capabilities and timing resemble M3.1-Flash-Preview. The evidence cannot distinguish M3.1 from another MiniMax build or special serving channel.
Is Space Bunny GLM?
The current evidence argues against that theory. Reported tokenizer probes match MiniMax much more strongly than GLM. This is attribution evidence, not an official identity statement.
What is Space Bunny's context window?
Community testing commonly reports about 1M tokens of context and about 524K maximum output. These are observed preview properties, not published API guarantees.
Is Space Bunny free?
It has been free during the preview on the named gateways. Free access can be rate-limited, interrupted or changed, and does not establish permanent pricing or retention terms.
Can I use Space Bunny on APIMaster?
No confirmed public Space Bunny route is available on APIMaster at this check. A callable MiniMax alternative is minimax-m3, whose live marketplace card should be checked for current price and route availability.
Is Space Bunny good for coding agents?
It can be effective for fast exploratory coding, bug fixes and iterative agent work. It is less dependable for strict formats, exact output lengths, simple tasks under high reasoning effort and workflows that require stable uptime or a documented model contract.
Does Space Bunny not train on prompts?
OpenRouter has described a no-training-on-prompts policy for its route, but gateway policies are not automatically identical. Check the current terms of the gateway you use and do not generalize one provider's policy to every Space Bunny endpoint.
Sources and current status
- OpenRouter — gateway where
stealth/space-bunny-alphawas reported; check the current model listing and terms before use. - OpenCode — coding-agent gateway where
space-bunny-freewas reported. - MiniMax — provider product pages and MiniMax Code access; MiniMax has not officially claimed Space Bunny in the evidence tracked here.
- APIMaster live MiniMax M3 card — current route count and price checked September 28, 2026.
- MiniMax M3.1 Flash Preview status page — separate tracker for the MiniMax Code preview and its public API status.
For the practical API path today, register with APIMaster, create an API key, set the OpenAI-compatible base URL to https://apimaster.ai/v1, and test minimax-m3. Return to this page when MiniMax or either gateway publishes a verifiable Space Bunny identity or API contract.