space-bunny-alphaSpace Bunny Alpha is a stealth-preview language model from a provider that has chosen to stay anonymous; it appeared on OpenRouter on 23 September 2026 as stealth/space-bunny-alpha and is available here as space-bunny-alpha on the OpenAI-compatible /v1/chat/completions endpoint. Its published capabilities are a 1M-token context window, up to 524K output tokens, adjustable reasoning effort (reasoning_effort, with the reasoning trace returned in the response), tool calling and tool_choice, response_format, and image and video input. Rather than repeat the provider's claims, we ran it through 28 tasks on 29 September 2026, one real API call each, and opened every result: 21 passed, 5 were partial, 2 failed. All six 3D and web tasks passed, from a Three.js solar system with orbit controls to a 64 KB Chart.js admin dashboard whose sorting, filtering, dark mode and collapsible sidebar all worked in Chrome with no console errors. Writing was 4 of 4: a seven-character quatrain that scans correctly line by line, a Chinese-to-English technical translation with every number intact, a 2,300-word engineering article and native-quality Japanese and Korean product copy. Reasoning was 3 of 3, including a deliberately under-constrained seating puzzle that it correctly reported as having four solutions instead of inventing one, and a three-day itinerary that met every budget, walking and attraction constraint. It retrieved 3 of 3 planted facts from a 91,298-token document with exact quotes, and made two parallel tool calls with correct arguments in 2 seconds. The four coding tasks were all partial for the same reason: the logic was right but the details were not checked. It missed one of the bugs in an async-pool function, wrote 13 tests for a correct LRU cache of which 3 assert the opposite of its own implementation, left a garbled token in an otherwise sound SQL query, and printed a hand-computed pandas table with 4 of 12 quantiles wrong. A 1,450-line canvas explainer failed on a single missing quote; with that character restored it rendered five scenes. Two advertised capabilities did not hold on this route: a strict JSON schema sent through response_format was accepted but not enforced (two required keys renamed, extra fields added), so validate on your side; and video input arrives as a few sampled frames without audio, so it can describe what is in a clip but not its length, motion or sound. Image input is good: it named Preikestolen over Lysefjord from a photo unprompted and estimated the crowd on the plateau correctly. Two settings matter more than anything else. First, reasoning_effort: at the default the model reasoned for 4 to 5 minutes before the first character on code, planning and even a four-line poem, and 4 of 28 calls were cut upstream around 300 seconds; at low or medium the same prompts answered in 4 to 40 seconds to first token with no loss we could see. Second, max_tokens: the reasoning trace is billed inside completion_tokens and streams before the answer, so a small cap returns empty content; omit it (the gateway forwards nothing and the upstream uses the model's own limit) or set it large. Throughput is roughly 90 to 140 tokens per second including reasoning. Pricing is $0.10 input / $0.40 output per 1M tokens, cached input $0.01: the model is free upstream during the preview and the price covers the gateway, so there is no official list price to compare against, and reasoning tokens count as output. It is a preview: stealth models are usually withdrawn or renamed once their provider reveals them, so treat the model id as temporary and keep a fallback in your own code. The provider's stealth terms say prompts and completions may be retained by them, though not used for training; do not send data you would not share with an unnamed third party. Rate limits are the preview's, not ours: sustained bursts can return 429, and failed calls are never billed.
View complete API reference with all parameters and examples.
View complete API reference with streaming, thinking, and more.
Billing: Cost = (input_tokens * input_price + output_tokens * output_price) / 1,000,000
A 1,000,000-token window and up to 524K output tokens, per the published endpoint metadata; 3 of 3 planted facts retrieved from a 91K-token document with exact quotes
At the default it reasoned 4 to 5 minutes before the first character and 4 of 28 calls were cut upstream; low or medium answered the same prompts in 4 to 40 seconds with no visible loss
Two parallel function calls with exact arguments in 2 seconds. response_format is accepted but a strict JSON schema was not enforced in our test: validate on your side
Named Preikestolen from a photo unprompted and described a generated character in detail. Video arrives as a few sampled frames without audio, so do not rely on it for duration or sound
$0.10 / $0.40 per 1M while the preview lasts. The provider is anonymous and may retain prompts and completions without training on them; expect the id to change when it is revealed
Space Bunny Alpha is a Large Language Model API provided by Stealth. Space Bunny Alpha is a stealth-preview language model from a provider that has chosen to stay anonymous; it appeared on OpenRouter on 23 September 2026 as stealth/space-bunny-alpha and is available here as space-bunny-alpha on the OpenAI-compatible /v1/chat/completions endpoint. Its published capabilities are a 1M-token context window, up to 524K output tokens, adjustable reasoning effort (reasoning_effort, with the reasoning trace returned in the response), tool calling and tool_choice, response_format, and image and video input. Rather than repeat the provider's claims, we ran it through 28 tasks on 29 September 2026, one real API call each, and opened every result: 21 passed, 5 were partial, 2 failed. All six 3D and web tasks passed, from a Three.js solar system with orbit controls to a 64 KB Chart.js admin dashboard whose sorting, filtering, dark mode and collapsible sidebar all worked in Chrome with no console errors. Writing was 4 of 4: a seven-character quatrain that scans correctly line by line, a Chinese-to-English technical translation with every number intact, a 2,300-word engineering article and native-quality Japanese and Korean product copy. Reasoning was 3 of 3, including a deliberately under-constrained seating puzzle that it correctly reported as having four solutions instead of inventing one, and a three-day itinerary that met every budget, walking and attraction constraint. It retrieved 3 of 3 planted facts from a 91,298-token document with exact quotes, and made two parallel tool calls with correct arguments in 2 seconds. The four coding tasks were all partial for the same reason: the logic was right but the details were not checked. It missed one of the bugs in an async-pool function, wrote 13 tests for a correct LRU cache of which 3 assert the opposite of its own implementation, left a garbled token in an otherwise sound SQL query, and printed a hand-computed pandas table with 4 of 12 quantiles wrong. A 1,450-line canvas explainer failed on a single missing quote; with that character restored it rendered five scenes. Two advertised capabilities did not hold on this route: a strict JSON schema sent through response_format was accepted but not enforced (two required keys renamed, extra fields added), so validate on your side; and video input arrives as a few sampled frames without audio, so it can describe what is in a clip but not its length, motion or sound. Image input is good: it named Preikestolen over Lysefjord from a photo unprompted and estimated the crowd on the plateau correctly. Two settings matter more than anything else. First, reasoning_effort: at the default the model reasoned for 4 to 5 minutes before the first character on code, planning and even a four-line poem, and 4 of 28 calls were cut upstream around 300 seconds; at low or medium the same prompts answered in 4 to 40 seconds to first token with no loss we could see. Second, max_tokens: the reasoning trace is billed inside completion_tokens and streams before the answer, so a small cap returns empty content; omit it (the gateway forwards nothing and the upstream uses the model's own limit) or set it large. Throughput is roughly 90 to 140 tokens per second including reasoning. Pricing is $0.10 input / $0.40 output per 1M tokens, cached input $0.01: the model is free upstream during the preview and the price covers the gateway, so there is no official list price to compare against, and reasoning tokens count as output. It is a preview: stealth models are usually withdrawn or renamed once their provider reveals them, so treat the model id as temporary and keep a fallback in your own code. The provider's stealth terms say prompts and completions may be retained by them, though not used for training; do not send data you would not share with an unnamed third party. Rate limits are the preview's, not ours: sustained bursts can return 429, and failed calls are never billed. Through APIMODELS platform, you can access this model via a unified API with transparent pay-as-you-go pricing. Current pricing: Input: $0.10, Output: $0.40 per 1M tokens.
Build intelligent conversational systems to automatically answer user queries and improve service efficiency.
Automatically write articles, emails, ad copy, and other text content to boost productivity.
Assist with code writing, debugging, and code review to accelerate software development.
Understand and analyze unstructured data, extract key insights, and generate summary reports.
Space Bunny Alpha is available through APIMODELS at: Input: $0.10, Output: $0.40 per 1M tokens. Billing is pay-as-you-go — you only pay for what you generate.
Sign up at APIMODELS, get your API key, and call our unified API endpoint. We provide detailed API documentation with code examples in cURL, Python, and Node.js.
APIMODELS offers the same Space Bunny Alpha model through our aggregation platform. We provide a unified API interface so you do not need separate accounts for each provider - one API key to access all models.
A stealth-preview language model whose provider has not identified itself. It appeared on OpenRouter on 23 September 2026 as stealth/space-bunny-alpha; on apimodels it is space-bunny-alpha on the OpenAI-compatible /v1/chat/completions endpoint. Published specs: a 1M-token context window, up to 524K output tokens, reasoning_effort, tool calling, response_format, image and video input. Stealth models are normally renamed or withdrawn once their maker reveals them, so treat the id as temporary and keep a fallback model in your code.
21 passed, 5 partial, 2 failed, on 29 September 2026, one real API call per task with every result opened. 3D and web pages 6 of 6 (Three.js scenes, a pure-CSS 3D page, a Chart.js dashboard, all working in Chrome with no console errors). Writing 4 of 4, including a seven-character Chinese quatrain with correct tones and a translation that kept every number. Reasoning 3 of 3; it even reported an under-constrained puzzle as having four solutions instead of inventing one. Long context 3 of 3 needles from a 91K-token document; tool calling perfect in 2 seconds. The four coding tasks were partial because it does not check its own output: one missed bug, three self-written tests that contradict its correct implementation, a garbled SQL token, four wrong hand-computed quantiles. A canvas animation failed on one missing quote.
Set reasoning_effort to low or medium for anything that generates a lot: code, pages, plans, long articles. At the default the model reasoned for 4 to 5 minutes before the first character and 4 of our 28 calls were cut upstream around 300 seconds; at low or medium the same prompts reached the first token in 4 to 40 seconds with no quality difference we could see. For max_tokens, omit it or set it large: the reasoning trace is billed inside completion_tokens and streams before the answer, so a small cap returns empty content. The gateway forwards nothing when you omit it, and the upstream applies the model's own 524K limit. Use streaming for long generations.
Not as a guarantee. In our test a strict JSON schema sent through response_format came back as valid JSON with the right facts, but two required keys were renamed and every line item carried an extra field despite additionalProperties: false. Nothing enforced the schema on this route, so validate the response on your side and retry on failure. Tool calling, by contrast, was clean: two parallel function calls with exact arguments, finish_reason tool_calls, in 2 seconds.
Images, yes, and well: from a landscape photo it named Preikestolen over Lysefjord unprompted, estimated the crowd on the plateau correctly and described a generated fantasy character in detail. Video arrives as a handful of sampled frames with no audio: it described the opening and closing frames of a 2.02-second clip accurately but estimated its length at 0.2 seconds and could not tell there was a soundtrack. Use video input for what is in a clip, not for duration, motion or sound.
$0.10 per 1M input tokens, $0.40 per 1M output and $0.01 for cached input, billed on actual usage; reasoning tokens count as output, failed calls are free. The model is free upstream during the preview and this price covers the gateway, so there is no official list price to compare. The provider's stealth terms say prompts and completions may be retained by them but are not used for training; do not send data you would not share with an unnamed third party. Rate limits belong to the preview: sustained bursts can return 429.
On APIMODELS, Space Bunny Alpha runs alongside 60+ models on one API key and one balance, so choosing is about fit, not lock-in. It supports Stealth Preview, 1M Context, Reasoning Effort, Tool Calling, Image + Video Input, and you can weigh it on price and capability against other Large Language Model models, then switch by changing a single model-name string — no new account or integration. Browse every Large Language Model option with live pricing at apimodels.app/models.
Space Bunny Alpha supports: Stealth Preview, 1M Context, Reasoning Effort, Tool Calling, Image + Video Input. See the APIMODELS docs for full parameters and call examples.
Yes. APIMODELS exposes Space Bunny Alpha through a single unified API and one key — no separate provider accounts, and no need to handle each provider's regional network access yourself.
We support Stripe (Visa, Mastercard, and other international cards) and Alipay. Credits are available instantly after payment.
How to get access, regional availability, and how this model compares with its alternatives.