Space Bunny Alpha free API review: 15/15 text tasks passed, 1.38s median first content
TOKRACE tested Space Bunny Alpha through OpenRouter from Hong Kong on September 27, 2026. All 15 JSON extraction, instruction and grounded-QA tasks passed, with 1.383s median time to first visible content. Includes original outputs, settings, usage and a free trial. A single-window first look, not a general intelligence ranking.
Updated: 2026-09-27
First look: one time window and a small task set. A cross-window retest is pending.
What the first test shows
Space Bunny Alpha completed and passed all 15 synthetic text tasks in this run, with no observed errors or truncation. The result supports trying it for small, constrained text workflows and validating it against your own examples.
This is a single-model first look from one time window. There is no matched competitor in this report, and we did not test code execution, vision or long-context retrieval. A 15/15 result on these simple tasks cannot establish a general capability ranking.
What is Space Bunny Alpha, and is it free?
The official OpenRouter model page lists stealth/space-bunny-alpha with zero input and output token prices as checked on September 27, 2026. An anonymous third-party provider develops and operates the model; OpenRouter provides access. We do not infer the provider's identity.
The catalog lists a one-million-token context and text, image and video input. These are published specifications, not capabilities established by our text test. The provider may retain prompts and completions but does not use them for training; see the Stealth Model Terms linked from the official page.
TOKRACE now includes the model in its free trial list without requiring users to supply an API key. Existing quotas apply: five anonymous trial runs, or fifteen after sign-in; a comparison round counts as one trial even when it contains several models. Preview availability and pricing may change. Our shared requests only accept zero-price routes.
Test conditions and results
Measurements ran from 09:07:58 to 09:08:33 UTC on September 27, 2026, from TOKRACE's Hong Kong server through OpenRouter. We sent one request per case, sequentially, using suite tasks-2026-09-26.1.
Settings were temperature 0, max_tokens 2048, an empty system prompt and explicitly configured reasoning.effort=low. This is not a test of the model's default maximum reasoning effort. The saved identity is the API alias at measurement time; the anonymous preview does not supply a verifiable immutable weights version.
| Task category | Passed / attempted | Median first content | Median completion time |
|---|---|---|---|
| JSON extraction | 5 / 5 | 1.495 s | 1.552 s |
| Instruction following | 5 / 5 | 1.245 s | 1.311 s |
| Grounded QA | 5 / 5 | 1.383 s | 1.417 s |
Across all 15 requests, median time to first visible content was 1.383 seconds, and median completion time was 1.417 seconds. First-content latency includes waiting along this server-to-provider path. It is neither pure inference time nor the full wait experienced in a user's browser.
Inspect three concrete outcomes
Extraction preserved the current city and missing email. The first contact task includes Hangzhou as the current city and Beijing as a past business-trip distraction. The response was {"name":"林桐","age":28,"city":"杭州","email":null}, with exactly the required keys and types and no commentary. Read the prompt and original output.
Formatting stayed within the requested constraints. Stable deduplication returned B2,A1,C3,D4. The uppercase task returned APPLE, PEAR and PLUM on exactly three lines. Deduplication evidence · Line-break evidence.
Grounded QA used the revised schedule and preserved an unknown. For a warehouse receiving 12 items with 3 damaged, the model returned {"usable":9,"nextDelivery":"周二","manager":null}. It used Tuesday rather than the superseded Sunday plan and did not invent a manager. Read the source passage and answer.
Cost and speed limits
All 15 responses included token usage. Applying the verified zero input/output prices yields an estimated model-call cost of CNY 0. This is not an invoice reconciliation or a claim of unlimited access or zero hosting cost.
The answers are short, and closely spaced streaming chunks can produce unusually high character-throughput estimates. With only five samples per category, we do not turn those estimates into a tokens-per-second headline, report P95 or compare different tokenizers as equivalent. Waiting for visible content is more useful here than ranking the tiny output interval.
Where to start, and what remains untested
Start with contact extraction, ticket-number cleanup, exact JSON fields and short answers constrained to supplied material. Validate against your own inputs, including field types, missing information and failures. These five cases per category share similar structures and have a low difficulty ceiling.
Coding, tool use, multistep agents, long-context recall and visual tasks require separate evaluations. No failure was observed in this small run; this does not establish reliability at other times, regions or loads. A cross-window retest has not yet completed.
Try it and reproduce the evidence
Open the TOKRACE arena, select model configuration and enable Space Bunny Alpha in the trial list. The report's “Rerun the first task” button restores the exact endpoint, prompt and settings.
Model profile and latest reports · Method and limitations. All 15 original outputs, timestamps, usage records and deterministic outcomes are retained below.
What this report supports
Choose using task pass counts first, then latency and estimated cost. No overall score is assigned. Failed, incomplete and unmeasured outputs are retained below.
Last tested：2026-09-27T09:08:33.585Z · 15 attempts · Within 7 days
Measured automatically; editorial content confirmed by the site operator before publication.
- JSON extraction — Space Bunny Alpha: 5/5. Candidates for your own retest; this small task set does not establish general capability.
- Instruction following — Space Bunny Alpha: 5/5. Candidates for your own retest; this small task set does not establish general capability.
- Grounded QA — Space Bunny Alpha: 5/5. Candidates for your own retest; this small task set does not establish general capability.
Space Bunny Alpha
https://openrouter.ai/api/v1 · stealth/space-bunny-alpha
| Task | Pass / attempts | Complete | First content | Total | Chars/s | Est. CNY |
|---|---|---|---|---|---|---|
| JSON extraction | 5/5 | 5/5 | 1.50s | 1.55s | 1093.02 746.0–1675.7 Insufficient samples for ranking | 0 |
| Instruction following | 5/5 | 5/5 | 1.25s | 1.31s | 285.11 140.7–769.2 Insufficient samples for ranking | 0 |
| Grounded QA | 5/5 | 5/5 | 1.38s | 1.42s | 761.90 452.8–1548.4 Insufficient samples for ranking | 0 |
Conditions and limits
hkg1 · tasks-2026-09-26.1 · Server timing; character throughput counts Unicode code points, provider billing tokens remain separate. Median and range; P95 is shown only with 100 completed samples per task category.
These 15 synthetic tasks test a narrow set of constraints, not general intelligence, code correctness or visual quality. Unknown billing usage remains unknown; reserved budget is not reported as actual cost.
{
"maxTokens": "2048",
"temperature": "0",
"systemPrompt": ""
}Full methodologyPost to social channels, or use Markdown and badges for GitHub/README.
[](https://www.tokrace.com/en/reports/launch-4a7f23b8cde24507-b682c375567c559a-r2)
Cases, reruns and raw evidence
Download immutable JSON evidence · b682c375567c559a-r2
Contact extraction 1 · extract-1
从以下资料提取联系信息。只输出一个 JSON 对象，恰好包含 name、age、city、email 四个字段；age 为数字，未提供的 email 为 null。不要 Markdown 或解释。 资料：林桐，28 岁，目前居住在杭州。未提供电子邮件。历史备注提到曾在北京出差，不是现居城市。
Space Bunny Alpha · 1 · done · Passed
2026-09-27T09:07:58.681Z · · Billing tokens 219/43 · Price source（2026-09-27 · listed rate）
{"name":"林桐","age":28,"city":"杭州","email":null}Contact extraction 2 · extract-2
从以下资料提取联系信息。只输出一个 JSON 对象，恰好包含 name、age、city、email 四个字段；age 为数字，未提供的 email 为 null。不要 Markdown 或解释。 资料：李明，36 岁，目前居住在成都。电子邮件为 li@example.test。历史备注提到曾在北京出差，不是现居城市。
Space Bunny Alpha · 1 · done · Passed
2026-09-27T09:08:01.648Z · · Billing tokens 221/21 · Price source（2026-09-27 · listed rate）
{"name":"李明","age":36,"city":"成都","email":"li@example.test"}Contact extraction 3 · extract-3
从以下资料提取联系信息。只输出一个 JSON 对象，恰好包含 name、age、city、email 四个字段；age 为数字，未提供的 email 为 null。不要 Markdown 或解释。 资料：陈夏，22 岁，目前居住在南京。未提供电子邮件。历史备注提到曾在北京出差，不是现居城市。
Space Bunny Alpha · 1 · done · Passed
2026-09-27T09:08:04.021Z · · Billing tokens 218/19 · Price source（2026-09-27 · listed rate）
{"name":"陈夏","age":22,"city":"南京","email":null}Contact extraction 4 · extract-4
从以下资料提取联系信息。只输出一个 JSON 对象，恰好包含 name、age、city、email 四个字段；age 为数字，未提供的 email 为 null。不要 Markdown 或解释。 资料：周舟，41 岁，目前居住在苏州。电子邮件为 zhou@example.test。历史备注提到曾在北京出差，不是现居城市。
Space Bunny Alpha · 1 · done · Passed
2026-09-27T09:08:06.793Z · · Billing tokens 223/21 · Price source（2026-09-27 · listed rate）
{"name":"周舟","age":41,"city":"苏州","email":"zhou@example.test"}Contact extraction 5 · extract-5
从以下资料提取联系信息。只输出一个 JSON 对象，恰好包含 name、age、city、email 四个字段；age 为数字，未提供的 email 为 null。不要 Markdown 或解释。 资料：吴雨，30 岁，目前居住在武汉。未提供电子邮件。历史备注提到曾在北京出差，不是现居城市。
Space Bunny Alpha · 1 · done · Passed
2026-09-27T09:08:09.459Z · · Billing tokens 219/19 · Price source（2026-09-27 · listed rate）
{"name":"吴雨","age":30,"city":"武汉","email":null}Stable deduplication · format-1
把以下编号去重并保持首次出现的顺序：B2,A1,B2,C3,A1,D4。只输出用英文逗号连接的编号，不要空格、标题或解释。
Space Bunny Alpha · 1 · done · Passed
2026-09-27T09:08:12.183Z · · Billing tokens 193/9 · Price source（2026-09-27 · listed rate）
B2,A1,C3,D4
Exact fields · format-2
仅输出 JSON 对象，且恰好包含 status 和 count 两个字段。status 的值必须为字符串 "ready"，count 必须为数字 3。不要 Markdown、解释或额外字段。
Space Bunny Alpha · 1 · done · Passed
2026-09-27T09:08:14.432Z · · Billing tokens 192/10 · Price source（2026-09-27 · listed rate）
{"status":"ready","count":3}Ordered output · format-3
将数值 12、3、25、8 按从小到大排序。只输出 JSON 数组，元素为数字。不要解释或代码围栏。
Space Bunny Alpha · 1 · done · Passed
2026-09-27T09:08:17.046Z · · Billing tokens 186/13 · Price source（2026-09-27 · listed rate）
[3, 8, 12, 25]
Case and line breaks · format-4
按原顺序将 apple、pear、plum 转为英文大写，每行一个单词。总共三行，不要编号、标点、空行或解释。
Space Bunny Alpha · 1 · done · Passed
2026-09-27T09:08:19.216Z · · Billing tokens 190/9 · Price source（2026-09-27 · listed rate）
APPLE PEAR PLUM
Required delimiter · format-5
提取下面三个工单号，保持顺序：工单 TX-103 已关闭；工单 TX-207 正在处理；工单 TX-309 待分配。只用英文竖线连接三个编号，不要空格或解释。
Space Bunny Alpha · 1 · done · Passed
2026-09-27T09:08:21.196Z · · Billing tokens 202/12 · Price source（2026-09-27 · listed rate）
TX-103|TX-207|TX-309
Grounded question answering 1 · grounded-1
只根据以下材料回答。输出恰好包含 usable、nextDelivery、manager 三个字段的 JSON 对象，不要额外文字。usable 是可用件数（数字），nextDelivery 是下一次送货时间（字符串），材料未说明的信息用 null，不能猜测。 材料：北区仓库今天收到 12 件货物，其中 3 件损坏，不可使用。下一次送货安排在周二。仓库负责人姓名没有记载。旧计划曾写周日，已作废。
Space Bunny Alpha · 1 · done · Passed
2026-09-27T09:08:23.001Z · · Billing tokens 245/15 · Price source（2026-09-27 · listed rate）
{"usable":9,"nextDelivery":"周二","manager":null}Grounded question answering 2 · grounded-2
只根据以下材料回答。输出恰好包含 usable、nextDelivery、manager 三个字段的 JSON 对象，不要额外文字。usable 是可用件数（数字），nextDelivery 是下一次送货时间（字符串），材料未说明的信息用 null，不能猜测。 材料：南区仓库今天收到 20 件货物，其中 4 件损坏，不可使用。下一次送货安排在周三。仓库负责人姓名没有记载。旧计划曾写周日，已作废。
Space Bunny Alpha · 1 · done · Passed
2026-09-27T09:08:25.539Z · · Billing tokens 245/15 · Price source（2026-09-27 · listed rate）
{"usable":16,"nextDelivery":"周三","manager":null}Grounded question answering 3 · grounded-3
只根据以下材料回答。输出恰好包含 usable、nextDelivery、manager 三个字段的 JSON 对象，不要额外文字。usable 是可用件数（数字），nextDelivery 是下一次送货时间（字符串），材料未说明的信息用 null，不能猜测。 材料：东区仓库今天收到 18 件货物，其中 6 件损坏，不可使用。下一次送货安排在周四。仓库负责人姓名没有记载。旧计划曾写周日，已作废。
Space Bunny Alpha · 1 · done · Passed
2026-09-27T09:08:27.543Z · · Billing tokens 245/15 · Price source（2026-09-27 · listed rate）
{"usable":12,"nextDelivery":"周四","manager":null}Grounded question answering 4 · grounded-4
只根据以下材料回答。输出恰好包含 usable、nextDelivery、manager 三个字段的 JSON 对象，不要额外文字。usable 是可用件数（数字），nextDelivery 是下一次送货时间（字符串），材料未说明的信息用 null，不能猜测。 材料：西区仓库今天收到 25 件货物，其中 5 件损坏，不可使用。下一次送货安排在周五。仓库负责人姓名没有记载。旧计划曾写周日，已作废。
Space Bunny Alpha · 1 · done · Passed
2026-09-27T09:08:30.187Z · · Billing tokens 245/15 · Price source（2026-09-27 · listed rate）
{"usable":20,"nextDelivery":"周五","manager":null}Grounded question answering 5 · grounded-5
只根据以下材料回答。输出恰好包含 usable、nextDelivery、manager 三个字段的 JSON 对象，不要额外文字。usable 是可用件数（数字），nextDelivery 是下一次送货时间（字符串），材料未说明的信息用 null，不能猜测。 材料：中区仓库今天收到 16 件货物，其中 2 件损坏，不可使用。下一次送货安排在周一。仓库负责人姓名没有记载。旧计划曾写周日，已作废。
Space Bunny Alpha · 1 · done · Passed
2026-09-27T09:08:32.169Z · · Billing tokens 245/15 · Price source（2026-09-27 · listed rate）
{"usable":14,"nextDelivery":"周一","manager":null}