Is deepseek-ai/deepseek-v4-flash genuine? Standard check result
Report ID MZZR2026091516582864A0B9Generated at 2026/09/16 00:58
Insufficient evidence to determine
Insufficient evidence to determine model identity. API, speed, tool, and token observations are recorded separately and do not establish identity.
Insufficient evidenceEvidence status
主要问题：不太听指令，让它怎么答它不照做。
EvaluatorMozhenzhen
Tested modelDeepSeek V4 Flash
ProvidersXiaohui Fan
Times tested19
Duration1 min 57 sec
Identity verdictInsufficient evidence to determine
Tested againstBasic capability observations (no calibrated reference)
Model reference evidence
No reference samples under matching conditions are available for this model.
Model self-identification and individual answers do not establish authenticity. Official references only reflect API behavior at the time of sampling.
Service performance score：92 / 100 · Not used to establish model identity
Rule version：2026-09-14.v1
Token usage
We show the usage the endpoint reported itself; estimates appear only when there is no real number.
Observed usage552.8KThis is the real usage the endpoint reported itself, and the main number this page goes by.Usage estimated from the questions1.7MWhen the vendor gives no reference, we estimate from the questions we asked.Variance assessment67% belowIt reported less usage than we estimated; this seller may count or compress differently.Usage conclusionLess than estimatedThis is only the usage the endpoint reported; what you actually pay is on the bill.
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.
Tested 5 times
Share billed at the cache rate36.3%
Low cache share
Caching was used, but only a little; check your bill to see how much it really saves.
Usage this run13.4KCached tokens4.9KNon-cached tokens8.5KAverage per run2.7K
Cache share by testThis is only the cache share the endpoint reported; the real discount is on the bill
Long-context test
We test how much it can read at once, then compare that with the officially stated length.
测了 1 档
Verified up to 1M
All tested context tiers (1M) returned reliably.
Reference windowReference window 1MModel-center context window; available lengths may vary by provider.Tested context lengths1MTests begin at the longest generated context and step down to verify stable, correct responses.
Test details
Expand any row for details; anything private has been removed.
14 checks
Model list visibility
deepseek-ai/deepseek-v4-flash is visible in the API model list.
Passed
Test details
这题测了 2 次，结果：Passed。
Basic call probe
The basic API call succeeded and returned a recognizable response.
Passed
Test details
这题测了 1 次，结果：Passed。
Model self-identification
Asked the model to identify itself; all 1 responses matched the expected name.
Passed
Test details
这题测了 1 次：1 次Passed。
本题用量 143 token
指令跟随
用几道题看它的回答水平像不像官方模型。 1 次里对了 0 次。
Needs review
Test details
这题测了 1 次：1 次要留意。
没Passed的原因：指令跟随不稳定
本题用量 33 token
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
这题测了 1 次：1 次Passed。
本题用量 32 token
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
这题测了 1 次：1 次Passed。
本题用量 86 token
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
这题测了 1 次：1 次Passed。
本题用量 58 token
品牌边界
Asked the model to identify itself; all 1 responses matched the expected name.
Passed
Test details
这题测了 1 次：1 次Passed。
本题用量 78 token
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
这题测了 1 次：1 次Passed。
本题用量 46 token
DeepSeek 推理边界
Asked the model to identify itself; all 1 responses matched the expected name.
Passed
Test details
这题测了 1 次：1 次Passed。
本题用量 107 token
模型代码锚定
Asked the model to identify itself; all 1 responses matched the expected name.
Passed
Test details
这题测了 1 次：1 次Passed。
本题用量 133 token
Model-family test: series and tier identification
Asked the model to identify itself; all 1 responses matched the expected name.
Insufficient evidence to determine model identity. API, speed, tool, and token observations are recorded separately and do not establish identity.
EvaluatorMozhenzhen
Tested modelDeepSeek V4 Flash
ProvidersXiaohui Fan
Times tested19
Duration2 min
Identity verdictInsufficient evidence to determine
Tested againstBasic capability observations (no calibrated reference)
Model reference evidence
No reference samples under matching conditions are available for this model.
Model self-identification and individual answers do not establish authenticity. Official references only reflect API behavior at the time of sampling.
Service performance score：92 / 100 · Not used to establish model identity
Rule version：2026-09-14.v1
Token usage
We show the usage the endpoint reported itself; estimates appear only when there is no real number.
Observed usage552.8KThis is the real usage the endpoint reported itself, and the main number this page goes by.Usage estimated from the questions1.7MWhen the vendor gives no reference, we estimate from the questions we asked.Variance assessment67% belowIt reported less usage than we estimated; this seller may count or compress differently.Usage conclusionLess than estimatedThis is only the usage the endpoint reported; what you actually pay is on the bill.
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.
Tested 5 times
Share billed at the cache rate36.3%
Low cache share
Caching was used, but only a little; check your bill to see how much it really saves.
Usage this run13.4KCached tokens4.9KNon-cached tokens8.5KAverage per run2.7K
Cache share by testThis is only the cache share the endpoint reported; the real discount is on the bill
Long-context test
We test how much it can read at once, then compare that with the officially stated length.
测了 1 档
Verified up to 1M
All tested context tiers (1M) returned reliably.
Reference windowReference window 1MModel-center context window; available lengths may vary by provider.Tested context lengths1MTests begin at the longest generated context and step down to verify stable, correct responses.
Test details
Expand any row for details; anything private has been removed
14 checks
Model list visibility
deepseek-ai/deepseek-v4-flash is visible in the API model list.
Passed
Test details
这题测了 2 次，结果：Passed。
Basic call probe
The basic API call succeeded and returned a recognizable response.
Passed
Test details
这题测了 1 次，结果：Passed。
Model self-identification
Asked the model to identify itself; all 1 responses matched the expected name.
Passed
Test details
这题测了 1 次：1 次Passed。
本题用量 143 token
指令跟随
用几道题看它的回答水平像不像官方模型。 1 次里对了 0 次。
Needs review
Test details
这题测了 1 次：1 次要留意。
没Passed的原因：指令跟随不稳定
本题用量 33 token
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
这题测了 1 次：1 次Passed。
本题用量 32 token
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
这题测了 1 次：1 次Passed。
本题用量 86 token
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
这题测了 1 次：1 次Passed。
本题用量 58 token
品牌边界
Asked the model to identify itself; all 1 responses matched the expected name.
Passed
Test details
这题测了 1 次：1 次Passed。
本题用量 78 token
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
这题测了 1 次：1 次Passed。
本题用量 46 token
DeepSeek 推理边界
Asked the model to identify itself; all 1 responses matched the expected name.
Passed
Test details
这题测了 1 次：1 次Passed。
本题用量 107 token
模型代码锚定
Asked the model to identify itself; all 1 responses matched the expected name.
Passed
Test details
这题测了 1 次：1 次Passed。
本题用量 133 token
Model-family test: series and tier identification
Asked the model to identify itself; all 1 responses matched the expected name.
