openai/gpt-5 · Model identity and integrity report

Last updated：Sep 13, 2026, 6:05 PM · Data window: past 90 days

Verification capability — fingerprint reference available

Family: openai
V3 sub-model reference: Available ✓
Checks: 知識截止 · 能力探針 · 拒絕模板 · 拒絕溫度參數
Family-fingerprint coverage: 52 weighted rules · 8 feature categories

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

Measured results (90 days)

4 runs · 4 hosts · 3 anomalies

Family-fingerprint consistency

  openai           3  ███████████████████░░░░░░   75% (expected)
  unknown          1  ██████░░░░░░░░░░░░░░░░░░░   25% (suspicious)

Observed V3 sub-model mapping

  → abstain                                  2 (50%) — low confidence
  → unknown                                  1 (25%) ⚠ mismatch
  → openai/gpt-5-mini                        1 (25%) ⚠ mismatch sim=0.76

Capability score distribution (auxiliary)

  0-49     ███████████████████░░░░░░    3
  50-69    ░░░░░░░░░░░░░░░░░░░░░░░░░    0
  70-89    ██████░░░░░░░░░░░░░░░░░░░    1
  90-100   ░░░░░░░░░░░░░░░░░░░░░░░░░    0

Host integrity ranking

Host	Runs	Average score	Family match	Most common V3 match
https://agonizing-reflex-darkened.ngrok-free.dev/v1	1	0	0%	—
https://alloy-liver-congressional-role.trycloudflare.com/v1	1	33	100%	—
https://conduit.ozdoev.net/v1	1	80	100%	openai/gpt-5-mini (1)
https://vip.j3gb.com/v1	1	0	100%	—

Recent anomalies

https://alloy-liver-congressional-role.trycloudflare.com/v1Jul 28, 2026, 5:11 AM

Self-claimed model: gpt-5
V2 predicted family: openai
V3 best match: —
Verdict: V3 low confidence · Abstain

https://conduit.ozdoev.net/v1Jul 11, 2026, 3:26 AM

Self-claimed model: gpt-5
V2 predicted family: openai
V3 best match: openai/gpt-5-mini (sim 0.76)
Verdict: Sub-model downgrade

https://vip.j3gb.com/v1Jul 9, 2026, 11:27 AM

Self-claimed model: gpt-5
V2 predicted family: openai
V3 best match: —
Verdict: V3 low confidence · Abstain

Methodology

BazaarLink Probe combines a Bayesian V2 family classifier across nine fingerprint categories with V3 sub-model references to verify endpoints claiming openai/gpt-5. Low family consistency, a different V3 match, or divergence from selfClaim is flagged as anomalous. BazaarLink Probe home.

Test an endpoint yourself →