Highly similarBelow confirmation threshold
与官方 kimi-k3 高度相似
行为特征与官方 kimi-k3 高度相似，未达确认标准，与 kimi-k3.1-pro 难以区分。
Brand consistency97/100
Model match score (kimi-k3)25/100
Checks13/17 Passed
- Providers
- 爱堡云AI
- Claimed model
- kimi-k3
- Plan
- Full check
- Reference
- 108 official models
- Requests
- 24
- Duration
- 8 分钟
01Identity check
Behaviour compared with each official referenceIndependent sample A采样 AValidIndependent sample B采样 BValidIndependent sample C采样 CValidIndependent sample D采样 DValidIndependent sample E采样 EValidIndependent sample F采样 FValidIndependent sample G采样 GValidIndependent sample H采样 HInvalidIndependent sample I采样 IValidReference108 official models · 8/9 Valid
Answer by answer8 answers: 4 match kimi-k3；another 1 look more like kimi-k3.1-pro；another 3 match no model clearly on their own
Similarity to official modelsTop 5
- kimi-k3.1-pro73%
- kimi-k3Claimed model25%
- claude-opus-5-5<1%
- gpt-5.4-mini<1%
- claude-fable-5-1<1%
02Test details
17 items · 13 passed · 4 notes · 0 failed| No. | Check | Result | Methodology |
|---|---|---|---|
| Connectivity & auth | |||
| A-01 | Endpoint reachable | Passed | The endpoint responds normally |
| A-02 | Key authentication | Passed | API key is valid and allowed to make calls |
| A-03 | Model on sale | Passed | The model list includes kimi-k3 |
| API conformance | |||
| B-01 | Basic call | Passed | Returns the expected content as instructed |
| B-02 | Response structure | Meets the spec | All OpenAI-compatible fields are present |
| B-03 | Model ID echoed back | Match | The model ID in the response matches the request |
| B-04 | Usage record | Complete and consistent | Input, output and total are all present, and the total equals the sum |
| B-05 | Output cleanliness | Nothing added | No inserted ads, watermarks or extra notes found |
| Identity check | |||
| C-01 | Independent sampling | 8 / 9 valid | 首轮未能判定，已追加两轮；部分采样结果不可用于比对 |
| C-02 | Brand consistency | Match | Behavior belongs to the Moonshot Kimi family; brand match score 97 |
| C-03 | Model consistency | Below confirmation threshold | 型号一致度 25，与 kimi-k3.1-pro 难以区分 |
| C-04 | Cross-brand swap check | None found | Similarity to other brands' models totals under 5% |
| Billing & capabilities | |||
| D-01 | Cache billing | Cache billing works | On average 98% of repeated requests were billed at the cache price. |
| D-02 | Context length | 400K passed | Read in full up to 400K; 1M failed: 接口返回 HTTP 504. |
| D-03 | Streaming output | Normal | Streamed content is complete |
| D-04 | Tool use | Supported | Returns the correct function name and arguments |
| D-05 | Structured output | Did not follow the schema | The response did not follow the required JSON schema |
03Which apps can connect
Tried once per app with kimi-k3, the way each app connectsUsing this test’s address, key and model, we test responses and tool calls in each app’s API format. We do not launch the apps or test installation, sign-in or complete tasks.
| Software | Result | Methodology |
|---|---|---|
| Codex | Connects | Replied normally and called the toolConnected the way Codex does (OpenAI Responses format) |
| Cherry Studio, Chatbox, LobeChat, Open WebUI, Cline | Connects | Replied normally and called the toolConnected the way Cherry Studio and similar apps do (OpenAI-compatible format) |
Cache billing check
The same long content was sent 5 times: the 1st is billed at the regular price, and from the 2nd on the repeated part should be billed at the cache price.
Cache billing ratio98.4%
Cache billing works
On average 98% of repeated requests were billed at the cache price.
Usage this run26.1KCached tokens20.5KNon-cached tokens5.7KAverage per run5.2K
Usage reportedReportedCache billingHappenedResultCache billing works
Cache share by testThis is only the cache share the endpoint reported; the real discount is on the bill
Context length check
Long documents are sent tier by tier to check whether the whole document is read.
Verified up to 400K
Read in full up to 400K; 1M failed: 接口返回 HTTP 504.
Reference windowVerified up to 400KThe model center has no reference context window for this model yet.Tested context lengths32K / 128K / 400K / 1MTested from short to long, stopping at the first failure.
- 32KRead in full · 27.0K received
- 128KRead in full · 121.8K received
- 400KRead in full · 376.0K received
- 1M接口返回 HTTP 504
Token usage check
Reported usage is used; estimates appear only when none is reported.
Reported usage564.3KThe actual usage reported by the API; this page goes by it.Estimated usage1.3M-1.8MEstimated from this test's requests.Deviation56% belowThe reported usage is below the estimate, possibly because of how tokens are counted or compressed.Usage checkBelow the estimateFigures are as reported by the API; your bill is authoritative.
Method: independent samples from the tested endpoint are compared against reference samples collected from official channels; billing, context and capability items come from real calls. API keys, request bodies and raw model output are not stored.
This report reflects the endpoint at the time of testing.
Reviews
还没有人评论，来写第一条。
Before you buy AI, check MozhenzhenScan to view reportOpen full report
Mozhenzhen verification report · Measured with real calls
kimi-k3 verification report
Highly similarBelow confirmation threshold
与官方 kimi-k3 高度相似
行为特征与官方 kimi-k3 高度相似，未达确认标准，与 kimi-k3.1-pro 难以区分。
Brand consistency97/100
Model match score (kimi-k3)25/100
Checks13/17 Passed
- Providers
- 爱堡云AI
- Claimed model
- kimi-k3
- Plan
- Full check
- Reference
- 108 official models
- Requests
- 24
- Duration
- 8 分钟
01Identity check
Behaviour compared with each official referenceIndependent sample A采样 AValidIndependent sample B采样 BValidIndependent sample C采样 CValidIndependent sample D采样 DValidIndependent sample E采样 EValidIndependent sample F采样 FValidIndependent sample G采样 GValidIndependent sample H采样 HInvalidIndependent sample I采样 IValidReference108 official models · 8/9 Valid
Answer by answer8 answers: 4 match kimi-k3；another 1 look more like kimi-k3.1-pro；another 3 match no model clearly on their own
Similarity to official modelsTop 5
- kimi-k3.1-pro73%
- kimi-k3Claimed model25%
- claude-opus-5-5<1%
- gpt-5.4-mini<1%
- claude-fable-5-1<1%
02Test details
17 items · 13 passed · 4 notes · 0 failed| No. | Check | Result | Methodology |
|---|---|---|---|
| Connectivity & auth | |||
| A-01 | Endpoint reachable | Passed | The endpoint responds normally |
| A-02 | Key authentication | Passed | API key is valid and allowed to make calls |
| A-03 | Model on sale | Passed | The model list includes kimi-k3 |
| API conformance | |||
| B-01 | Basic call | Passed | Returns the expected content as instructed |
| B-02 | Response structure | Meets the spec | All OpenAI-compatible fields are present |
| B-03 | Model ID echoed back | Match | The model ID in the response matches the request |
| B-04 | Usage record | Complete and consistent | Input, output and total are all present, and the total equals the sum |
| B-05 | Output cleanliness | Nothing added | No inserted ads, watermarks or extra notes found |
| Identity check | |||
| C-01 | Independent sampling | 8 / 9 valid | 首轮未能判定，已追加两轮；部分采样结果不可用于比对 |
| C-02 | Brand consistency | Match | Behavior belongs to the Moonshot Kimi family; brand match score 97 |
| C-03 | Model consistency | Below confirmation threshold | 型号一致度 25，与 kimi-k3.1-pro 难以区分 |
| C-04 | Cross-brand swap check | None found | Similarity to other brands' models totals under 5% |
| Billing & capabilities | |||
| D-01 | Cache billing | Cache billing works | On average 98% of repeated requests were billed at the cache price. |
| D-02 | Context length | 400K passed | Read in full up to 400K; 1M failed: 接口返回 HTTP 504. |
| D-03 | Streaming output | Normal | Streamed content is complete |
| D-04 | Tool use | Supported | Returns the correct function name and arguments |
| D-05 | Structured output | Did not follow the schema | The response did not follow the required JSON schema |
03Which apps can connect
Tried once per app with kimi-k3, the way each app connectsUsing this test’s address, key and model, we test responses and tool calls in each app’s API format. We do not launch the apps or test installation, sign-in or complete tasks.
| Software | Result | Methodology |
|---|---|---|
| Codex | Connects | Replied normally and called the toolConnected the way Codex does (OpenAI Responses format) |
| Cherry Studio, Chatbox, LobeChat, Open WebUI, Cline | Connects | Replied normally and called the toolConnected the way Cherry Studio and similar apps do (OpenAI-compatible format) |
Cache billing check
The same long content was sent 5 times: the 1st is billed at the regular price, and from the 2nd on the repeated part should be billed at the cache price.
Cache billing ratio98.4%
Cache billing works
On average 98% of repeated requests were billed at the cache price.
Usage this run26.1KCached tokens20.5KNon-cached tokens5.7KAverage per run5.2K
Usage reportedReportedCache billingHappenedResultCache billing works
Cache share by testThis is only the cache share the endpoint reported; the real discount is on the bill
Context length check
Long documents are sent tier by tier to check whether the whole document is read.
Verified up to 400K
Read in full up to 400K; 1M failed: 接口返回 HTTP 504.
Reference windowVerified up to 400KThe model center has no reference context window for this model yet.Tested context lengths32K / 128K / 400K / 1MTested from short to long, stopping at the first failure.
- 32KRead in full · 27.0K received
- 128KRead in full · 121.8K received
- 400KRead in full · 376.0K received
- 1M接口返回 HTTP 504
Token usage check
Reported usage is used; estimates appear only when none is reported.
Reported usage564.3KThe actual usage reported by the API; this page goes by it.Estimated usage1.3M-1.8MEstimated from this test's requests.Deviation56% belowThe reported usage is below the estimate, possibly because of how tokens are counted or compressed.Usage checkBelow the estimateFigures are as reported by the API; your bill is authoritative.
Method: independent samples from the tested endpoint are compared against reference samples collected from official channels; billing, context and capability items come from real calls. API keys, request bodies and raw model output are not stored.
This report reflects the endpoint at the time of testing.
Reviews
还没有人评论，来写第一条。
发表评论
编辑器加载中…
发表评论