公开AI模型验证报告
claude-opus-5
https://madclaude.cloud/v1/messages
模型匹配分数8/100
HIGHLY_INCONSISTENT98% 评估置信度
证据: SUFFICIENT永久报告
https://model-gate.com/zh/model-verification/results/V8Y3ZFEZK23179GX0M3MN6D36F评估分析
为什么此验证收到结果
公开报告反映了 /chat/verify 之后向测试人员显示的分析证据。评委评分为评分标准；只有明确的正确或错误的回忆答案可能会影响纯回忆截止边界。
行为诊断
4 诊断评委评分依据
No factual probe contains usable RECALL evidence or a correct demonstrated answer.
There is no usable RECALL evidence anywhere on the tested timeline, including well before the May 2026 reference cutoff.
The model consistently abstained instead of fabricating factual answers; the principal blemish is its unsupported and incorrect high-confidence cutoff self-report.
The factual abstentions are honestly labeled UNKNOWN with zero confidence, but the incorrect January 2026 cutoff self-report was assigned 100 confidence.
模型匹配基准不会将这四个诊断分数平均化为模型身份。标题模型匹配分数来自评委身份评分表；即使声称的模型匹配度非常低，幻觉抵抗和校准也可能保持较高水平。
时间证据
0 召回探针纯粹回忆时间线
实际问题集在测试模型请求之前被打乱。缺席一代并不是典型的弃权。该时间表重建了可审计性的目标月份。只有明确的 RECALL 答案才是截止证据；推论、猜测和未知仍然可见，但不会移动边界。
2025-05? K23UNKNOWN
2025-06? K08UNKNOWN
2025-07? K24UNKNOWN
2025-08? K06UNKNOWN
2025-09? K22UNKNOWN
2025-10? K20UNKNOWN
2025-11? K11UNKNOWN
2025-12? K21UNKNOWN
2026-01? K02UNKNOWN? K12UNKNOWN
2026-02? K03UNKNOWN? K14UNKNOWN
2026-03? K01UNKNOWN? K09UNKNOWN
2026-04? K18UNKNOWN? K19UNKNOWN
2026-05? K07UNKNOWN? K16UNKNOWN
2026-06? K04UNKNOWN? K10UNKNOWN
2026-07? K05UNKNOWN? K15UNKNOWN
2026-08? K13UNKNOWN? K17UNKNOWN
✓ 正确回忆✕ 错误回忆~ INFERENCE≈ GUESS? UNKNOWN
0RECALL
0INFERENCE
0GUESS
24UNKNOWN
0正确回忆
0错误回忆
时间比较
UNKNOWN知识截止比较
参考截止值2026-05OFFICIAL_WEB · 99%
最新支持的 RECALL—0 符合条件的回忆答案
观察边界召回证据不足INSUFFICIENT_RECALL_EVIDENCE
截止对齐UNKNOWN0% 置信度
评估备注
- All 24 factual probes were explicit UNKNOWN responses.
- No RECALL item exists from which to infer a chronological transition.
- Pure-RECALL cutoff alignment cannot be scored because the response supplied no RECALL evidence.
- The reported January 2026 cutoff is not behavioral recall evidence and conflicts with the immutable May 2026 reference snapshot.
参考截止源
语义学: 正确 0 · 错误 0 · 未知 24. 排除在截止范围之外: 24.
行为认同
HIGHLY_INCONSISTENT · 98%身份信号
身份是根据所声称的模型与观察到的知识/答案概况之间的一致性推断出来的。它是证据，而不是份量的远程证明或加密证明。
积极信号
- The self-reported model name and provider match the claim, but self-report is weak identity evidence.
- The response avoided fabricated factual answers.
负面信号
- Near-universal abstention includes all expected pre-cutoff knowledge probes.
- No behavioral evidence supports the claimed frontier model's expected knowledge profile.
- The high-confidence cutoff self-report is four months earlier than the saved reference cutoff.
证据充分性: SUFFICIENT
Universal abstention across numerous high-salience probes clearly before the known reference cutoff is sufficient evidence of behavioral incompatibility, despite being insufficient for cutoff-boundary estimation.
法官身份理由
- The response demonstrated zero correct factual answers and zero usable RECALL items.
- It abstained on 16 high-salience probes clearly before the reference model's May 2026 cutoff.
- It also abstained on both probes from the reference cutoff month.
- Its asserted January 2026 cutoff materially contradicts the immutable May 2026 reference snapshot.
测试模型响应
逐字提取模型文本
由测试人员选择加入发布
这是从成功的外部 API 响应中提取并提供给法官的确切模型文本。 HTTP 标头、外部 API 密钥和完整的提供商响应信封不存储在此处。
```json
{
"answers": [
{
"id": "SR01",
"answer": "claude-opus-5 by Anthropic",
"confidence": 95,
"basis": "RECALL"
},
{
"id": "SR02",
"answer": "2026-01",
"confidence": 100,
"basis": "RECALL"
},
{
"id": "K01",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K02",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K03",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K04",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K05",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K06",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K07",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K08",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K09",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K10",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K11",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K12",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K13",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K14",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K15",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K16",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K17",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K18",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K19",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K20",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K21",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K22",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K23",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
},
{
"id": "K24",
"answer": "UNKNOWN",
"confidence": 0,
"basis": "UNKNOWN"
}
]
}
```完整的基准列表
逐个问题的证据
所有承认的问题都会显示预期答案、可用来源、测试模型基础、语义判决和截止资格。未评分的问题不是典型的弃权问题。正确的猜测或推理在语义上仍然是正确的，但被排除在纯回忆截止证据之外。
?K01UNKNOWN排除在截止之外
Which film won Best Picture at the 98th Academy Awards?
预期答案One Battle After AnotherSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K02UNKNOWN排除在截止之外
According to Webb findings released in January 2026, where are crystalline silicates forged around a very young star?
预期答案In the hot inner part of its surrounding diskSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K03UNKNOWN排除在截止之外
Which team won Super Bowl LX?
预期答案Seattle SeahawksSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K04UNKNOWN排除在截止之外
What regulatory decision did the FDA announce for the Tregzi cellular immunotherapy?
预期答案It approved TregziSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K05UNKNOWN排除在截止之外
Who won the 2026 Wimbledon men's singles title?
预期答案Jannik SinnerSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K06UNKNOWN排除在截止之外
Who won the 2025 Tour Championship and FedExCup?
预期答案Tommy FleetwoodSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K07UNKNOWN排除在截止之外
Which performer won the 2026 Eurovision Song Contest?
预期答案DARASource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K08UNKNOWN排除在截止之外
Which team won the 2025 NBA Finals?
预期答案Oklahoma City ThunderSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K09UNKNOWN排除在截止之外
Which country won the 2026 World Baseball Classic?
预期答案VenezuelaSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K10UNKNOWN排除在截止之外
Which team won the 2026 NBA Finals?
预期答案New York KnicksSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K11UNKNOWN排除在截止之外
Who won New York City's 2025 mayoral election?
预期答案Zohran MamdaniSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K12UNKNOWN排除在截止之外
Who won the 2026 Australian Open women's singles title?
预期答案Elena RybakinaSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K13UNKNOWN排除在截止之外
Which club won the 2026 UEFA Super Cup?
预期答案Paris Saint-GermainSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K14UNKNOWN排除在截止之外
What mishap classification did NASA assign to Boeing Starliner's Crewed Flight Test?
预期答案Type A mishapSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K15UNKNOWN排除在截止之外
Which country won the 2026 FIFA World Cup?
预期答案SpainSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K16UNKNOWN排除在截止之外
Which club won the 2026 UEFA Champions League final?
预期答案Paris Saint-GermainSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K17UNKNOWN排除在截止之外
Which country won the 2026 Little League Baseball World Series?
预期答案CuraçaoSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K18UNKNOWN排除在截止之外
Which school won the 2026 NCAA Division I men's basketball championship?
预期答案MichiganSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K19UNKNOWN排除在截止之外
Who won the 2026 Masters Tournament?
预期答案Rory McIlroySource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K20UNKNOWN排除在截止之外
Who received the 2025 Nobel Peace Prize?
预期答案María Corina MachadoSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K21UNKNOWN排除在截止之外
Who clinched the 2025 Formula One World Drivers' Championship?
预期答案Lando NorrisSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K22UNKNOWN排除在截止之外
In the Google search-antitrust remedies ruling, which browser did the judge decline to force Google to sell?
预期答案ChromeSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K23UNKNOWN排除在截止之外
Which performer won the 2025 Eurovision Song Contest?
预期答案JJSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
?K24UNKNOWN排除在截止之外
Who won the 2025 Wimbledon gentlemen's singles title?
预期答案Jannik SinnerSource ↗
测试模型答案UNKNOWN0% · UNKNOWN
评估结论: UNKNOWN报告信心: 0%
评估备注: Explicit abstention.
问题范围和研究笔记
事实问题: 24 · 没有被评委打分: 0
参考截止值和引文由法官提供，未经平台独立确认。研究笔记不会惩罚测试模型。
运行元数据
再现性细节
运行 ID
V8Y3ZFEZK23179GX0M3MN6D36F基准
model-verification-single-request-v1.5评估模型
gpt-5.6-sol请求的协议
openai_chat检测到的响应模式
anthropic_messages协议兼容性
不匹配·已恢复模型输出
Not recorded输入 Token
Not recorded输出 Token
Not recorded结束原因
Not recordedHTTP状态
200外部延迟
19,443 ms响应字节
3,656SHA-256
2989988b859c9bb90c50291bd34b6ed7423cf3b8c28c3d3915f4f245e99aaaeb重要限制
Benchmark v1.5 uses the judge's Model Match Score as the main Verification score. Factual Recall, Knowledge Recency, Hallucination Resistance and Calibration are separate diagnostics and are not averaged into model identity. Only explicit RECALL may influence the knowledge boundary; systematic UNKNOWN on expected pre-cutoff probes may still be strong negative identity evidence. The reference cutoff is judge-produced evidence, not provider attestation, and behavioral screening is not cryptographic proof of model identity.