grok45 · Model identity and integrity report

Family

xai

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.

658 runs · 319 hosts · 569 anomalies

Family-fingerprint consistency

  xai            376  ██████████████░░░░░░░░░░░   57% (suspicious)
  anthropic      130  █████░░░░░░░░░░░░░░░░░░░░   20% (suspicious)
  unknown         89  ████░░░░░░░░░░░░░░░░░░░░░   14% (suspicious)
  openai          62  ██░░░░░░░░░░░░░░░░░░░░░░░    9% (suspicious)
  google           1  ░░░░░░░░░░░░░░░░░░░░░░░░░    0% (suspicious)

Observed V3 sub-model mapping

  → abstain                                190 (29%) — low confidence
  → x-ai/grok-4.5                          167 (25%) ⚠ mismatch sim=0.76
  → x-ai/grok-4.3                          167 (25%) ⚠ mismatch sim=0.76
  → unknown                                125 (19%) ⚠ mismatch
  → openai/gpt-5                             6 ( 1%) ⚠ mismatch sim=0.83
  → anthropic/claude-fable-5                 2 ( 0%) ⚠ mismatch sim=0.90
  → openai/gpt-5-mini                        1 ( 0%) ⚠ mismatch sim=0.79

Capability score distribution (auxiliary)

  0-49     ██████████████░░░░░░░░░░░  373
  50-69    ░░░░░░░░░░░░░░░░░░░░░░░░░    6
  70-89    █████████░░░░░░░░░░░░░░░░  227
  90-100   ██░░░░░░░░░░░░░░░░░░░░░░░   52

BazaarLink Probe combines a Bayesian V2 family classifier across nine fingerprint categories with V3 sub-model references to verify endpoints claiming grok45. Low family consistency, a different V3 match, or divergence from selfClaim is flagged as anomalous. BazaarLink Probe home.