glm51 · Model identity and integrity report

Family

zhipu

V3 sub-model reference

Available ✓

Checks

知識截止 · 能力探針 · 拒絕模板 · 拒絕溫度參數

Concrete reference answers, refusal templates, and individual signal weights are not published because disclosure would help spoofing endpoints evade detection.

88 runs · 52 hosts · 85 anomalies

Family-fingerprint consistency

  zhipu           44  █████████████░░░░░░░░░░░░   50% (suspicious)
  openai          21  ██████░░░░░░░░░░░░░░░░░░░   24% (suspicious)
  anthropic       15  ████░░░░░░░░░░░░░░░░░░░░░   17% (suspicious)
  google           4  █░░░░░░░░░░░░░░░░░░░░░░░░    5% (suspicious)
  unknown          3  █░░░░░░░░░░░░░░░░░░░░░░░░    3% (suspicious)
  deepseek         1  ░░░░░░░░░░░░░░░░░░░░░░░░░    1% (suspicious)

Observed V3 sub-model mapping

  → abstain                                 53 (60%) — low confidence
  → z-ai/glm-5                              15 (17%) ⚠ mismatch sim=0.79
  → z-ai/glm-5.2                             8 ( 9%) ⚠ mismatch sim=0.82
  → unknown                                  5 ( 6%) ⚠ mismatch
  → z-ai/glm-5.1                             3 ( 3%) ⚠ mismatch sim=0.84
  → deepseek/deepseek-v4-pro                 1 ( 1%) ⚠ mismatch sim=0.78
  → google/gemini-3.1-flash-lite-preview     1 ( 1%) ⚠ mismatch sim=0.73
  → anthropic/claude-opus-4.7                1 ( 1%) ⚠ mismatch sim=0.69
  → google/gemini-3.1-pro-preview            1 ( 1%) ⚠ mismatch sim=0.96

Capability score distribution (auxiliary)

  0-49     █████████████░░░░░░░░░░░░   44
  50-69    █░░░░░░░░░░░░░░░░░░░░░░░░    3
  70-89    ███████████░░░░░░░░░░░░░░   37
  90-100   █░░░░░░░░░░░░░░░░░░░░░░░░    4

BazaarLink Probe combines a Bayesian V2 family classifier across nine fingerprint categories with V3 sub-model references to verify endpoints claiming glm51. Low family consistency, a different V3 match, or divergence from selfClaim is flagged as anomalous. BazaarLink Probe home.