Space Bunny is useful precisely because it is easy to try: a free, 1M-context model with text, image, and video input. It is also a poor model ID to hardcode into a production app. The developer is still anonymous, and neither public route promises when the free preview ends.
This is a practical companion to our Space Bunny builder's guide. It focuses on the first request, what to measure, and how to make the eventual switch boring. This is not a benchmark or a claim that we ran the prompts below.
Check the route before you call it
As of September 29, 2026, the two published routes are:
| Route | Model ID | What the listing says |
|---|---|---|
| OpenRouter | stealth/space-bunny-alpha |
Free, 1M-token context, text/image/video input, text output, adjustable reasoning effort |
| OpenCode Zen |
space-bunny-free (opencode/space-bunny-free in OpenCode config) |
Free model in Zen's OpenAI-compatible chat-completions endpoint |
Both require an API key from the route you choose. Their data terms are different: OpenRouter says the anonymous third-party provider may retain prompts and completions but does not use them for training. OpenCode publishes its own Zen privacy policy and exceptions. Read the current terms before sending private code or customer data. A listing also does not guarantee that a request will succeed at any particular moment.
Make one small request
For OpenRouter, put your own key in OPENROUTER_API_KEY and send a low-stakes prompt:
curl https://openrouter.ai/api/v1/chat/completions \
-H "Authorization: Bearer $OPENROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "stealth/space-bunny-alpha",
"reasoning": { "effort": "low" },
"messages": [
{ "role": "user", "content": "Summarize this public changelog in five bullets: ..." }
]
}'
~~~{% endraw %}
For OpenCode Zen, use its own key and model ID at {% raw %}`https://opencode.ai/zen/v1/chat/completions`{% endraw %}. You cannot swap the two keys. In OpenCode's configuration, the model ID takes the {% raw %}`opencode/` prefix; in the direct Zen API request, the model ID is `space-bunny-free`.
Start with public, non-sensitive material. Check the returned model, response text, usage, latency, and any error before putting this into an agent loop. OpenRouter lists tool calling and JSON output support, but a feature flag is only a reason to test those paths, not a reliability guarantee.
## Build a tiny evaluation set before the preview disappears
The 1M context window makes Space Bunny interesting for repository navigation, long documents, and agent sessions. A single impressive demo will not tell you whether it works for *your* workload. A repeatable check is more useful:
1. Save 15–30 prompts from work you actually do, with a note about what a good answer must contain.
2. Include short, long-context, image, and tool-use cases only if your product uses them.
3. Run the same prompts through Space Bunny and two named models you would be willing to pay for.
4. Record output quality, source fidelity, latency, reasoning-token use, tool-call errors, and timeouts. Keep the full prompt and route with each result.
5. Re-run the set when the model ID, provider, price, or data terms change.
If you want examples of **publicly reported** Space Bunny tests and their limitations, see our separate [Hugging Face case roundup](https://huggingface.co/blog/karmen-beatapi/space-bunny-tested-real-cases-evaluation). We did not run those third-party tests ourselves.
## Keep an exit door in the code
Use one configuration value for the primary model and one for a named fallback:
~~~js
export const models = {
primary: process.env.PRIMARY_MODEL ?? 'stealth/space-bunny-alpha',
fallback: process.env.FALLBACK_MODEL ?? 'your-named-baseline',
};
~~~
Then decide the switch rule *now*: for example, move to the cheapest named model that stays within an acceptable quality gap on your own evaluation set when Space Bunny stops being free or its route becomes unreliable. Recheck pricing at that time; do not assume today's $0 listing will persist.
The full [Space Bunny builder's guide](https://spacebunny.im/blog/what-is-space-bunny/) keeps the route comparison, setup details, source links, and five-step exit plan in one place. It is an independent guide, not the model's official site.
*Disclosure: We publish spacebunny.im and also work on BeatAPI. BeatAPI does not provide Space Bunny access as of September 29, 2026. The model IDs and pricing above come from the linked OpenRouter and OpenCode listings, not from our own provider.*
Top comments (0)