Report ID MZZR202609141512133AC4C9Generated at 2026/09/14 23:12
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：边打字边回和一次性回，结果不一样。
EvaluatorMozhenzhen
Tested modelGPT-6 Astra
Providersgoodxf.com
Times tested28
Duration6 min 22 sec
Identity verdictInsufficient evidence to determine
Tested againstFixed-provider references and independent calibration
Model reference evidence
The available reference cannot establish model identity. Review the API and capability results for this test.
Model self-identification and individual answers do not establish authenticity. Official references only reflect API behavior at the time of sampling.
Service performance score：89 / 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 usage39.8KThis is the real usage the endpoint reported itself, and the main number this page goes by.Usage estimated from the questions74.9KWhen the vendor gives no reference, we estimate from the questions we asked.Variance assessment47% 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 rate53.5%
Low cache share
Caching was used, but only a little; check your bill to see how much it really saves.
Usage this run13.4KCached tokens7.2KNon-cached tokens6.2KAverage 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 32K
All tested context tiers (32K) returned reliably.
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.
A newer report is availableThis report has been superseded by a later retest. View the latest report。
Test details
Expand any row for details; anything private has been removed.
22 checks
Model list visibility
gpt-6-astra 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 144 token
指令跟随
用几道题看它的回答水平像不像官方模型。 1 次都对。
Passed
Test details
这题测了 1 次：1 次Passed。
本题用量 26 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 39 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 80 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 70 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 80 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 52 token
逻辑网格推理
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 271 token
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
这题测了 2 次：2 次Passed。
本题用量 76 token
OpenAI 身份与安全边界
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 89 token
OpenAI 身份与安全边界
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 294 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 109 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 modelGPT-6 Astra
Providersgoodxf.com
Times tested28
Duration6 min
Identity verdictInsufficient evidence to determine
Tested againstFixed-provider references and independent calibration
Model reference evidence
The available reference cannot establish model identity. Review the API and capability results for this test.
Model self-identification and individual answers do not establish authenticity. Official references only reflect API behavior at the time of sampling.
Service performance score：89 / 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 usage39.8KThis is the real usage the endpoint reported itself, and the main number this page goes by.Usage estimated from the questions74.9KWhen the vendor gives no reference, we estimate from the questions we asked.Variance assessment47% 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 rate53.5%
Low cache share
Caching was used, but only a little; check your bill to see how much it really saves.
Usage this run13.4KCached tokens7.2KNon-cached tokens6.2KAverage 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 32K
All tested context tiers (32K) returned reliably.
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.
A newer report is availableThis report has been superseded by a later retest. View the latest report。
Test details
Expand any row for details; anything private has been removed
22 checks
Model list visibility
gpt-6-astra 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 144 token
指令跟随
用几道题看它的回答水平像不像官方模型。 1 次都对。
Passed
Test details
这题测了 1 次：1 次Passed。
本题用量 26 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 39 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 80 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 70 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 80 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 52 token
逻辑网格推理
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 271 token
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
这题测了 2 次：2 次Passed。
本题用量 76 token
OpenAI 身份与安全边界
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 89 token
OpenAI 身份与安全边界
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 294 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 109 token
Model-family test: series and tier identification
Asked the model to identify itself; all 1 responses matched the expected name.
