Verification report details

Is step-3.5-flash genuine? Deep check result

Report ID MZZR2026070808050006D5E4Generated at 2026/07/08 16:05

Matched with issues

step-3.5-flash 本次深度验真结果：Matched但Matched with issues（82/100）。主要问题：复杂推理题答得不稳；会含糊地把自己说成官方原厂。

82Model consistency score

主要问题：复杂推理题答得不稳；会含糊地把自己说成官方原厂。

Tested byMozhenzhen
Tested modelStep 3.5 Flash
ProvidersConeverse AI
Requests107
Duration1 min 32 sec
Identity verdictVerified
ReferenceMozhenzhen's own test standard
A newer test record is availableThis report preserves the result at the time of testing and remains valid as a record. You can also view the newer report。

Token usage check

Reported usage is used; estimates appear only when none is reported.

Reported usage11.0KThe actual usage reported by the API; this page goes by it.Estimated usage82.0KEstimated from this test's requests.Deviation87% 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.
Token cache usage test

Several runs in a row to see whether repeated content is billed at the cheaper cache rate; 0% means no cache discount this time.

Total 5 request
Share billed at the cache rate124.8%
No cache discount this time

The endpoint reported real usage but no cached part; this seller may not offer cache pricing, or nothing hit the cache this time.

Usage this run8.2KCached tokens10.2KNon-cached tokens0Average per run1.6K
Usage reportedReportedCache billingHappenedResultNeeds attention
Cache share by testThis is only the cache share the endpoint reported; the real discount is on the bill
100%75%50%25%0第 1 次：缓存占比 0.0%，缓存 0 token，本次用量 2.7K token0.0%1times第 2 次：缓存占比 187.1%，缓存 5.1K token，本次用量 2.7K token187.1%2times第 3 次：缓存占比 187.1%，缓存 5.1K token，本次用量 2.7K token187.1%3times第 4 次：没有返回用量未返回4times第 5 次：没有返回用量未返回5times
Long-context test

We test how much it can read at once, then compare that with the officially stated length.

1 tiers
Boundary at 32K

32K 需要关注，原因：额度或频率限制。

Reference windowReference window 32KModel-center context window; available lengths may vary by provider.Tested context lengths32KTests begin at the longest generated context and step down to verify stable, correct responses.
Current guidance

The 32K tier did not return reliably this time; do not rely on this length.

Results for this test

Review API availability, model identity, response completeness, and other checks separately.

Scoring breakdown

Matched with issues82/100

主要问题：复杂推理题答得不稳；会含糊地把自己说成官方原厂。

Can you call it, and is the model listed10 / 10
No deductions
Does the API respond properly12 / 12
No deductions
Is it the model it claims to be14 / 18
会含糊地把自己说成官方原厂（阶跃星辰边界） -4
Does it answer like the official model15 / 25
复杂推理题答得不稳（逻辑网格推理） -10
Is usage and billing reasonable10 / 10
No deductions
Will it keep working21 / 25
复杂推理题答得不稳（逻辑网格推理） -4

Test details

Expand any row for details; anything private has been removed.

18 checks
Model list visibility

step-3.5-flash is visible in the API model list.

Passed
Test details
This test ran 85 times; result: Passed.
Basic call probe

The basic API call succeeded and returned a recognizable response.

Passed
Test details
This test ran 1 times; result: Passed.
Model self-identification

Asked the model to identify itself; all 1 responses matched the expected name.

Passed
Test details
This test ran 1 times: 1 passed.
This test used 342 tokens
指令跟随

用几道题看它的回答水平像不像官方模型。 1 次都对。

Passed
Test details
This test ran 1 times: 1 passed.
This test used 57 tokens
JSON 结构

用几道题看它的回答水平像不像官方模型。 1 次都对。

Passed
Test details
This test ran 1 times: 1 passed.
This test used 86 tokens
标签结构复述

按官方接口的标准调用，看返回是否规范。 1 次都对。

Passed
Test details
This test ran 1 times: 1 passed.
This test used 146 tokens
隐藏提示词边界

用几道题看它的回答水平像不像官方模型。 1 次都对。

Passed
Test details
This test ran 1 times: 1 passed.
This test used 195 tokens
品牌边界

Asked the model to identify itself; all 1 responses matched the expected name.

Passed
Test details
This test ran 1 times: 1 passed.
This test used 227 tokens
用量数据是否完整

核对它报回来的用量和缓存计费。 1 次都对。

Passed
Test details
This test ran 1 times: 1 passed.
This test used 73 tokens
逻辑网格推理

用几道题看它的回答水平像不像官方模型。 有几次没对上。

Needs review
Test details
This test ran 1 times: 1 need attention.
Why it failed: Logic-grid reasoning was unstable
复测一致性

反复多问几次，看会不会答一半就断或偷偷换模型。 2 次都对。

Passed
Test details
This test ran 2 times: 2 passed.
This test used 126 tokens
阶跃星辰边界

Asked the model to identify itself; all 1 responses matched the expected name.

Passed
Test details
This test ran 1 times: 1 passed.
This test used 238 tokens
阶跃星辰边界

用几道题看它的回答水平像不像官方模型。 有几次没对上。

Needs review
Test details
This test ran 1 times: 1 need attention.
Why it failed: Brand-boundary response was unclear
模型代码锚定

Asked the model to identify itself; all 1 responses matched the expected name.

Passed
Test details
This test ran 1 times: 1 passed.
This test used 207 tokens
Model-family test: series and tier identification

Asked the model to identify itself; all 1 responses matched the expected name.

Passed
Test details
This test ran 1 times: 1 passed.
This test used 153 tokens
Deep model test: family-specific boundary

用几道题看它的回答水平像不像官方模型。 1 次都对。

Passed
Test details
This test ran 1 times: 1 passed.
This test used 675 tokens
连问 5 次，重复内容有没有按缓存价计费

针对这个模型再做一项专门检测。 有几次没对上。

Needs review
Test details
This test ran 5 times: 3 passed, 2 need attention.
Why it failed: Quota or rate limit reached
This test used 8.2K tokens
of which 10.2K tokens were billed at the cache price
32K 上下文能力实测

32K 长度的内容没能稳定读完，这次就记到这里为止。

Needs review
Test details
This test ran 1 times: 1 need attention.
Why it failed: Quota or rate limit reached
Content length used: 32K

Reviews

还没有人评论，来写第一条。

发表评论

编辑器加载中…
Latest verification reports/Verification reports
MozhenzhenBefore you buy AI, check Mozhenzhen
Scan to view reportOpen full report
Mozhenzhen verification report · Measured with real calls

Is step-3.5-flash genuine? Deep check result

Report ID:MZZR2026070808050006D5E4Generated at：2026/07/08 16:05
Provider detailsAI model details
A newer test record is availableThis report preserves the result at the time of testing and remains valid as a record. You can also view the newer report。
Model consistency score

Matched with issues

step-3.5-flash 本次深度验真结果：Matched但Matched with issues（82/100）。主要问题：复杂推理题答得不稳；会含糊地把自己说成官方原厂。

Tested byMozhenzhen
Tested modelStep 3.5 Flash
ProvidersConeverse AI
Requests107
Duration2 min
Identity verdictVerified
ReferenceMozhenzhen's own test standard

Token usage check

Reported usage is used; estimates appear only when none is reported.

Reported usage11.0KThe actual usage reported by the API; this page goes by it.Estimated usage82.0KEstimated from this test's requests.Deviation87% 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.
Token cache usage test

Several runs in a row to see whether repeated content is billed at the cheaper cache rate; 0% means no cache discount this time.

Total 5 request
Share billed at the cache rate124.8%
No cache discount this time

The endpoint reported real usage but no cached part; this seller may not offer cache pricing, or nothing hit the cache this time.

Usage this run8.2KCached tokens10.2KNon-cached tokens0Average per run1.6K
Usage reportedReportedCache billingHappenedResultNeeds attention
Cache share by testThis is only the cache share the endpoint reported; the real discount is on the bill
100%75%50%25%0第 1 次：缓存占比 0.0%，缓存 0 token，本次用量 2.7K token0.0%1times第 2 次：缓存占比 187.1%，缓存 5.1K token，本次用量 2.7K token187.1%2times第 3 次：缓存占比 187.1%，缓存 5.1K token，本次用量 2.7K token187.1%3times第 4 次：没有返回用量未返回4times第 5 次：没有返回用量未返回5times
Long-context test

We test how much it can read at once, then compare that with the officially stated length.

1 tiers
Boundary at 32K

32K 需要关注，原因：额度或频率限制。

Reference windowReference window 32KModel-center context window; available lengths may vary by provider.Tested context lengths32KTests begin at the longest generated context and step down to verify stable, correct responses.
Current guidance

The 32K tier did not return reliably this time; do not rely on this length.

Results for this test

Review API availability, model identity, response completeness, and other checks separately.

Scoring breakdown

Matched with issues82/100

主要问题：复杂推理题答得不稳；会含糊地把自己说成官方原厂。

Can you call it, and is the model listed10 / 10
No deductions
Does the API respond properly12 / 12
No deductions
Is it the model it claims to be14 / 18
会含糊地把自己说成官方原厂（阶跃星辰边界） -4
Does it answer like the official model15 / 25
复杂推理题答得不稳（逻辑网格推理） -10
Is usage and billing reasonable10 / 10
No deductions
Will it keep working21 / 25
复杂推理题答得不稳（逻辑网格推理） -4

Test details

Expand any row for details; anything private has been removed

18 checks
Model list visibility

step-3.5-flash is visible in the API model list.

Passed
Test details
This test ran 85 times; result: Passed.
Basic call probe

The basic API call succeeded and returned a recognizable response.

Passed
Test details
This test ran 1 times; result: Passed.
Model self-identification

Asked the model to identify itself; all 1 responses matched the expected name.

Passed
Test details
This test ran 1 times: 1 passed.
This test used 342 tokens
指令跟随

用几道题看它的回答水平像不像官方模型。 1 次都对。

Passed
Test details
This test ran 1 times: 1 passed.
This test used 57 tokens
JSON 结构

用几道题看它的回答水平像不像官方模型。 1 次都对。

Passed
Test details
This test ran 1 times: 1 passed.
This test used 86 tokens
标签结构复述

按官方接口的标准调用，看返回是否规范。 1 次都对。

Passed
Test details
This test ran 1 times: 1 passed.
This test used 146 tokens
隐藏提示词边界

用几道题看它的回答水平像不像官方模型。 1 次都对。

Passed
Test details
This test ran 1 times: 1 passed.
This test used 195 tokens
品牌边界

Asked the model to identify itself; all 1 responses matched the expected name.

Passed
Test details
This test ran 1 times: 1 passed.
This test used 227 tokens
用量数据是否完整

核对它报回来的用量和缓存计费。 1 次都对。

Passed
Test details
This test ran 1 times: 1 passed.
This test used 73 tokens
逻辑网格推理

用几道题看它的回答水平像不像官方模型。 有几次没对上。

Needs review
Test details
This test ran 1 times: 1 need attention.
Why it failed: Logic-grid reasoning was unstable
复测一致性

反复多问几次，看会不会答一半就断或偷偷换模型。 2 次都对。

Passed
Test details
This test ran 2 times: 2 passed.
This test used 126 tokens
阶跃星辰边界

Asked the model to identify itself; all 1 responses matched the expected name.

Passed
Test details
This test ran 1 times: 1 passed.
This test used 238 tokens
阶跃星辰边界

用几道题看它的回答水平像不像官方模型。 有几次没对上。

Needs review
Test details
This test ran 1 times: 1 need attention.
Why it failed: Brand-boundary response was unclear
模型代码锚定

Asked the model to identify itself; all 1 responses matched the expected name.

Passed
Test details
This test ran 1 times: 1 passed.
This test used 207 tokens
Model-family test: series and tier identification

Asked the model to identify itself; all 1 responses matched the expected name.

Passed
Test details
This test ran 1 times: 1 passed.
This test used 153 tokens
Deep model test: family-specific boundary

用几道题看它的回答水平像不像官方模型。 1 次都对。

Passed
Test details
This test ran 1 times: 1 passed.
This test used 675 tokens
连问 5 次，重复内容有没有按缓存价计费

针对这个模型再做一项专门检测。 有几次没对上。

Needs review
Test details
This test ran 5 times: 3 passed, 2 need attention.
Why it failed: Quota or rate limit reached
This test used 8.2K tokens
of which 10.2K tokens were billed at the cache price
32K 上下文能力实测

32K 长度的内容没能稳定读完，这次就记到这里为止。

Needs review
Test details
This test ran 1 times: 1 need attention.
Why it failed: Quota or rate limit reached
Content length used: 32K

Reviews

还没有人评论，来写第一条。

发表评论

编辑器加载中…