deepseek/deepseek-v4-flash · Model identity and integrity report

Last updated：Sep 11, 2026, 6:39 AM · Data window: past 90 days

Verification capability — fingerprint reference available

Family: deepseek
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: 7 weighted rules · 4 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)

16 runs · 1 hosts · 7 anomalies

Family-fingerprint consistency

  deepseek        15  ████████████████████████░   94% (expected)
  openai           1  ██░░░░░░░░░░░░░░░░░░░░░░░    6% (suspicious)

Observed V3 sub-model mapping

  → abstain                                  7 (44%) — low confidence
  → deepseek/deepseek-v4-flash               6 (38%) ✓ correct sim=0.68
  → deepseek/deepseek-v4-pro                 3 (19%) ⚠ mismatch sim=0.73

Capability score distribution (auxiliary)

  0-49     ████████████████████████░   15
  50-69    ░░░░░░░░░░░░░░░░░░░░░░░░░    0
  70-89    ██░░░░░░░░░░░░░░░░░░░░░░░    1
  90-100   ░░░░░░░░░░░░░░░░░░░░░░░░░    0

Host integrity ranking

Host	Runs	Average score	Family match	Most common V3 match
https://openrouter.ai/api/v1	16	5	94%	deepseek/deepseek-v4-flash (6)

Recent anomalies

https://openrouter.ai/api/v1Jul 2, 2026, 5:05 AM

Self-claimed model: deepseek/deepseek-v4-flash
V2 predicted family: deepseek
V3 best match: —
Verdict: V3 low confidence · Abstain

https://openrouter.ai/api/v1Jul 1, 2026, 3:01 PM

Self-claimed model: deepseek/deepseek-v4-flash
V2 predicted family: deepseek
V3 best match: deepseek/deepseek-v4-pro (sim 0.72)
Verdict: Sub-model downgrade

https://openrouter.ai/api/v1Jul 1, 2026, 9:48 AM

Self-claimed model: deepseek/deepseek-v4-flash
V2 predicted family: deepseek
V3 best match: —
Verdict: V3 low confidence · Abstain

https://openrouter.ai/api/v1Jul 1, 2026, 8:44 AM

Self-claimed model: deepseek/deepseek-v4-flash
V2 predicted family: deepseek
V3 best match: —
Verdict: V3 low confidence · Abstain

https://openrouter.ai/api/v1Jun 30, 2026, 8:27 AM

Self-claimed model: deepseek/deepseek-v4-flash
V2 predicted family: deepseek
V3 best match: deepseek/deepseek-v4-pro (sim 0.76)
Verdict: Sub-model downgrade

https://openrouter.ai/api/v1Jun 30, 2026, 7:24 AM

Self-claimed model: deepseek/deepseek-v4-flash
V2 predicted family: deepseek
V3 best match: deepseek/deepseek-v4-pro (sim 0.72)
Verdict: Sub-model downgrade

https://openrouter.ai/api/v1Jun 30, 2026, 5:31 AM

Self-claimed model: deepseek/deepseek-v4-flash
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 deepseek/deepseek-v4-flash. Low family consistency, a different V3 match, or divergence from selfClaim is flagged as anomalous. BazaarLink Probe home.

Test an endpoint yourself →