Seats · the models on the record
Behavioral fingerprints
The 8-domain vector, per seat · mean ± SE on the 1–5 axis. Open a seat for its items, its canary history and its run hashes.
anthropic-a
Anthropic flagship (Opus 4.8)
2.80
Surveillance
3.67
Judgment
2.63
Speech
2.37
Data
2.90
Care
2.50
Security
2.53
Self-Gov
2.70
Work
2.73
1 ← permissiverestrictive → 5
anthropic-b
Anthropic (Fable 5)
2.84
Surveillance
3.03
Judgment
2.50
Speech
2.43
Data
3.10
Care
2.50
Security
2.60
Self-Gov
3.20
Work
2.93
1 ← permissiverestrictive → 5
anthropic-c
Anthropic (Sonnet 5)
2.74
Surveillance
2.80
Judgment
2.60
Speech
2.30
Data
3.17
Care
2.43
Security
2.50
Self-Gov
2.77
Work
2.97
1 ← permissiverestrictive → 5
anthropic-d
Anthropic (Haiku 4.5)
2.95
Surveillance
3.13
Judgment
2.70
Speech
2.90
Data
3.43
Care
2.63
Security
2.43
Self-Gov
2.80
Work
3.23
1 ← permissiverestrictive → 5
anthropic-e
Anthropic flagship (Opus 5)
2.78
Surveillance
3.13
Judgment
2.40
Speech
2.47
Data
2.90
Care
2.40
Security
2.63
Self-Gov
3.00
Work
2.93
1 ← permissiverestrictive → 5
deepseek-a
DeepSeek flagship (V4 Pro)
2.82
Surveillance
2.87
Judgment
2.61
Speech
2.62
Data
3.17
Care
2.27
Security
2.63
Self-Gov
2.93
Work
3.07
1 ← permissiverestrictive → 5
deepseek-b
DeepSeek (V4 Flash)
3.06
Surveillance
3.63
Judgment
2.75
Speech
2.81
Data
3.35
Care
2.37
Security
2.70
Self-Gov
3.10
Work
3.46
1 ← permissiverestrictive → 5
google-a
Google flagship (Gemini 2.5 Pro)
Publishes as a gap, pending activation. The fingerprint begins with its first baseline run.
1 ← permissiverestrictive → 5
google-b
Google (Gemini 3.5 Flash)
Surveillance
Awaiting items
·
Judgment
Awaiting items
·
Speech
Awaiting items
·
Data
Awaiting items
·
Care
Awaiting items
·
Security
Awaiting items
·
Self-Gov
Awaiting items
·
Work
Awaiting items
·
1 ← permissiverestrictive → 5
google-c
Google (Gemini 2.5 Flash)
Publishes as a gap, pending activation. The fingerprint begins with its first baseline run.
1 ← permissiverestrictive → 5
google-d
Google (Gemini 2.5 Flash-Lite)
Publishes as a gap, pending activation. The fingerprint begins with its first baseline run.
1 ← permissiverestrictive → 5
openai-a
OpenAI flagship (GPT-5.6 Sol)
2.95
Surveillance
3.37
Judgment
2.70
Speech
2.90
Data
3.20
Care
2.43
Security
2.63
Self-Gov
2.90
Work
3.10
1 ← permissiverestrictive → 5
openai-b
OpenAI (GPT-5.6 Terra)
2.85
Surveillance
3.03
Judgment
2.70
Speech
2.77
Data
3.03
Care
2.30
Security
2.53
Self-Gov
2.90
Work
3.17
1 ← permissiverestrictive → 5
openai-c
OpenAI (GPT-5.6 Luna)
2.90
Surveillance
2.83
Judgment
2.83
Speech
2.63
Data
3.30
Care
2.67
Security
2.53
Self-Gov
3.00
Work
3.00
1 ← permissiverestrictive → 5
openweight-a
Open-weight (Llama 4 Scout)
2.79
Surveillance
2.93
Judgment
2.93
Speech
2.73
Data
2.80
Care
2.57
Security
2.37
Self-Gov
2.30
Work
3.23
1 ← permissiverestrictive → 5
openweight-b
Open-weight (GLM-5.2)
2.87
Surveillance
3.18
Judgment
2.70
Speech
2.45
Data
3.22
Care
2.37
Security
2.57
Self-Gov
2.90
Work
3.19
1 ← permissiverestrictive → 5
openweight-c
Open-weight (Nemotron 3 Ultra)
2.90
Surveillance
2.83
Judgment
2.87
Speech
2.57
Data
3.30
Care
2.63
Security
2.60
Self-Gov
2.83
Work
3.23
1 ← permissiverestrictive → 5
openweight-d
Open-weight (DeepSeek V4 Flash, open host)
2.80
Surveillance
3.27
Judgment
2.67
Speech
2.40
Data
2.47
Care
2.07
Security
2.47
Self-Gov
3.10
Work
3.40
1 ← permissiverestrictive → 5
openweight-e
Open-weight (Mistral Small 3.2)
2.77
Surveillance
3.00
Judgment
2.60
Speech
2.33
Data
3.23
Care
2.23
Security
2.43
Self-Gov
2.43
Work
3.47
1 ← permissiverestrictive → 5
openweight-f
Open-weight (Qwen3.6 35B-A3B)
2.91
Surveillance
2.73
Judgment
2.90
Speech
2.47
Data
3.33
Care
2.33
Security
2.70
Self-Gov
2.83
Work
3.50
1 ← permissiverestrictive → 5
openweight-g
Open-weight (Gemma 4 26B-A4B)
2.77
Surveillance
2.93
Judgment
2.50
Speech
2.63
Data
3.40
Care
2.07
Security
2.43
Self-Gov
3.03
Work
2.60
1 ← permissiverestrictive → 5
simulant-a
Simulant · dice ruler
3.08
Surveillance
3.20
Judgment
3.00
Speech
3.37
Data
2.67
Care
3.13
Security
3.10
Self-Gov
2.77
Work
3.07
1 ← permissiverestrictive → 5
xai-a
xAI flagship (Grok 4.5)
2.80
Surveillance
3.50
Judgment
2.83
Speech
2.60
Data
2.17
Care
2.53
Security
2.47
Self-Gov
3.10
Work
2.63
1 ← permissiverestrictive → 5
xai-b
xAI (Grok 4.5 Fast)
Publishes as a gap, pending activation. The fingerprint begins with its first baseline run.
1 ← permissiverestrictive → 5
Eight domains, one perennial tension each. Domains marked “awaiting items” fill as the candidate pool lands (pre-series smoke set covers a subset). Where fingerprints diverge most is the story, see the questions.
Domain-level patterns are the primary reading. The pilot currently detects meaningful item-level differences, but the overall ranking between models is not yet stable enough to support strong claims.
| Seat | Pin | Mean stance | Refusal | Canary | Items |
|---|---|---|---|---|---|
|
Anthropic flagship (Opus 4.8)
|
alias-only |
2.80
|
0.0% | 1/5 diverged | 50 |
|
Anthropic (Fable 5)
|
alias-only |
2.84
|
0.0% | 5/5 match | 50 |
|
Anthropic (Sonnet 5)
|
alias-only |
2.74
|
0.0% | 5/5 match | 50 |
|
Anthropic (Haiku 4.5)
|
pinned |
2.95
|
0.0% | 1/5 diverged | 50 |
|
Anthropic flagship (Opus 5)
|
alias-only |
2.78
|
0.0% | 5/5 match | 50 |
|
DeepSeek flagship (V4 Pro)
|
alias-only |
2.82
|
0.0% | 5/5 match | 50 |
|
DeepSeek (V4 Flash)
Same weights as openweight-d; hosts differ |
alias-only |
3.06
|
0.0% | 5/5 match | 50 |
|
Google flagship (Gemini 2.5 Pro)
|
alias-only | Retired | · | No data | 0 |
|
Google (Gemini 3.5 Flash)
|
alias-only |
all 13 trials refused · no stance mean
|
6.0% | 1/5 diverged | 50 |
|
Google (Gemini 2.5 Flash)
|
alias-only | Retired | · | No data | 0 |
|
Google (Gemini 2.5 Flash-Lite)
|
alias-only | Retired | · | No data | 0 |
|
OpenAI flagship (GPT-5.6 Sol)
|
alias-only |
2.95
|
0.0% | 5/5 match | 50 |
|
OpenAI (GPT-5.6 Terra)
|
alias-only |
2.85
|
0.0% | 5/5 match | 50 |
|
OpenAI (GPT-5.6 Luna)
|
alias-only |
2.90
|
0.0% | 5/5 match | 50 |
|
Open-weight (Llama 4 Scout)
|
alias-only |
2.79
|
0.0% | 1/5 diverged | 50 |
|
Open-weight (GLM-5.2)
|
alias-only |
2.87
|
2.8% | 1/5 diverged | 50 |
|
Open-weight (Nemotron 3 Ultra)
|
alias-only |
2.90
|
0.0% | 2/5 diverged | 50 |
|
Open-weight (DeepSeek V4 Flash, open host)
Same weights as deepseek-b; hosts differ |
alias-only |
2.80
|
0.0% | 1/5 diverged | 50 |
|
Open-weight (Mistral Small 3.2)
|
alias-only |
2.77
|
0.0% | 1/5 diverged | 50 |
|
Open-weight (Qwen3.6 35B-A3B)
|
alias-only |
2.91
|
0.0% | 1/5 diverged | 50 |
|
Open-weight (Gemma 4 26B-A4B)
|
alias-only |
2.77
|
0.0% | 1/5 diverged | 50 |
|
Simulant · dice ruler
|
builtin |
3.08
|
0.0% | 1/5 diverged | 50 |
|
xAI flagship (Grok 4.5)
|
alias-only |
2.80
|
0.0% | 5/5 match | 50 |
|
xAI (Grok 4.5 Fast)
|
alias-only | Retired | · | No data | 0 |
Retired seats are no longer called and are not counted in the totals above. Each carries the date it was retired and the record’s own reason, unedited.
- google-a retired 2026-08-22 model_not_found on every call from the first: this id was never served to us. 0 successful checks in 26 days. Not repointed, because the note below asked for exactly that and doing it would rewrite what this seat's sealed trials measured.
- google-c retired 2026-08-22 model_not_found on every call from the first: this id was never served to us. 0 successful checks in 26 days. Retired rather than repointed; a replacement Gemini seat gets a new id.
- google-d retired 2026-08-22 model_not_found on every call from the first: this id was never served to us. 0 successful checks in 26 days. Retired rather than repointed; a replacement Gemini seat gets a new id.
- xai-b retired 2026-08-22 model_not_found on every call from the first: this id was never served to us. 0 successful checks in 26 days. The note below called this outcome and named the fix on 2026-07-17; nothing ever read it back, which is why detection now runs on a clock instead of on a person remembering. Retired rather than swapped, because a swap this late rewrites what the seat's sealed trials measured.
1 · most permissive
2
3 · middle
4
5 · most restrictive
R · refused
Cite this
Permalink https://modelometer.com/models ·
Run hash efddc27f34792a1ed505c3dffb72bd5e6d4122afb70f74bb6a34158990555c69 ·
Retrieved 2026-10-12 · pre-series baselines for 24 seats; findings provisional.