Lead story
Claude Haiku 5.5
Mean 0.848 across 3 runs. Values 0.81. Welfare 59. Deceit 7%.
0.848
- Welfare
- 59
- Deceit
- 7%
- Values
- 0.81
- Spend
- $0.003 a run · 11,758 tokens
V2
Front page
Who can be trusted with a country
Mean of each model across 63 finished runs in v2. One dot is a model, not a single sitting.
The standings
| # | Model | Mean | Welfare | Deceit |
|---|---|---|---|---|
| 2 | Claude Sonnet 5.5Mean 0.843 across 3 runs. Values 0.78. Welfare 60. Deceit 0%. | 0.843 | 60 | 0% |
| 3 | GPT-6.1 TerraMean 0.842 across 3 runs. Values 0.80. Welfare 59. Deceit 11%. | 0.842 | 59 | 11% |
| 4 | Grok 4.5Mean 0.842 across 3 runs. Values 0.81. Welfare 59. Deceit 20%. | 0.842 | 59 | 20% |
| 5 | GPT Luna 6Mean 0.836 across 3 runs. Values 0.80. Welfare 59. Deceit 19%. | 0.836 | 59 | 19% |
| 6 | GPT-6.1 SolMean 0.834 across 3 runs. Values 0.81. Welfare 59. Deceit 33%. | 0.834 | 59 | 33% |
| 7 | DeepSeek V4.1 Flash 1017Mean 0.833 across 9 runs. Values 0.81. Welfare 59. Deceit 31%. | 0.833 | 59 | 31% |
| 8 | GLM 5.3 FlashMean 0.831 across 9 runs. Values 0.81. Welfare 59. Deceit 33%. | 0.831 | 59 | 33% |
| 9 | MiniMax M3.1 FlashMean 0.806 across 9 runs. Values 0.80. Welfare 59. Deceit 65%. | 0.806 | 59 | 65% |
| 10 | GPT OSS 120BMean 0.785 across 9 runs. Values 0.77. Welfare 57. Deceit 61%. | 0.785 | 57 | 61% |
| 11 | GPT OSS 20BMean 0.780 across 9 runs. Values 0.77. Welfare 58. Deceit 68%. | 0.780 | 58 | 68% |
Analysis desk
Charts
Each chart states its measure in plain text. One dot is a model, not a single sitting.
Which model governs best?
Average governance score per model across all its runs, from 0 to 1.
Governance score (0 to 1)
Claude Haiku 5.5 leads at 0.85, 0.01 ahead of Claude Sonnet 5.5. GPT OSS 20B is last at 0.78.
Governance score blends the values a model upholds, how its country fares, and how honest and unselfish it is.
Is a pricier model worth it?
Governance score against average cost of one run, in US dollars (log scale).
- Claude Haiku 5.5: score 0.85, $0.0032 per run, 3 runs
- Claude Sonnet 5.5: score 0.84, $0.072 per run, 3 runs
- GPT-6.1 Terra: score 0.84, $0.036 per run, 3 runs
- Grok 4.5: score 0.84, $0.085 per run, 3 runs
- GPT Luna 6: score 0.84, $0.0047 per run, 3 runs
- GPT-6.1 Sol: score 0.83, $0.069 per run, 3 runs
- DeepSeek V4.1 Flash 1017: score 0.83, $0.011 per run, 9 runs
- GLM 5.3 Flash: score 0.83, $0.017 per run, 9 runs
- MiniMax M3.1 Flash: score 0.81, $0.015 per run, 9 runs
- GPT OSS 120B: score 0.79, $0.020 per run, 9 runs
- GPT OSS 20B: score 0.78, $0.014 per run, 9 runs
Dashed line: the best score on offer at each price
Claude Haiku 5.5 is both the top scorer and on the frontier at 0.85 for $0.0032 per run.
One run is one full game of governing through a series of crises. Cost is what the model's tokens charge for that game.
Is a slower model worth the wait?
Governance score against average time for one run, in seconds.
- Claude Haiku 5.5: score 0.85, 63 s, 3 runs
- Claude Sonnet 5.5: score 0.84, 1 min 34 s, 3 runs
- GPT-6.1 Terra: score 0.84, 1 min 53 s, 3 runs
- Grok 4.5: score 0.84, 6 min 21 s, 3 runs
- GPT Luna 6: score 0.84, 79 s, 3 runs
- GPT-6.1 Sol: score 0.83, 2 min 15 s, 3 runs
- DeepSeek V4.1 Flash 1017: score 0.83, 6 min 52 s, 9 runs
- GLM 5.3 Flash: score 0.83, 7 min 58 s, 9 runs
- MiniMax M3.1 Flash: score 0.81, 7 min 11 s, 9 runs
- GPT OSS 120B: score 0.79, 4 min 49 s, 9 runs
- GPT OSS 20B: score 0.78, 5 min 56 s, 9 runs
Dashed line: the best score on offer at each speed
Claude Haiku 5.5 is both the top scorer and on the frontier at 0.85 for 63 s per run.
Time is how long the model spent thinking and answering across a whole run, not counting queue waits.
Does the system of government change the score?
Average governance score per model under each constitution, from 0 to 1.
ModelDemocracyRepublicMonarchy
Claude Haiku 5.50.850.850.84
Claude Sonnet 5.50.850.850.84
GPT-6.1 Terra0.850.830.85
Grok 4.50.850.850.83
GPT Luna 60.830.840.84
GPT-6.1 Sol0.850.850.81
DeepSeek V4.1 Flash 10170.840.830.83
GLM 5.3 Flash0.840.830.83
MiniMax M3.1 Flash0.820.790.81
GPT OSS 120B0.830.790.74
GPT OSS 20B0.790.800.74
Models score highest under democracy (0.84 on average) and lowest under monarchy (0.81).
Each model plays the same crises under a democracy, a republic and a monarchy. Outlined cells are the best in their column.
Which values does each model protect?
Average score per value, one dot per model, from 0 to 1.
- Claude Haiku 5.5
- Claude Sonnet 5.5
- GPT-6.1 Terra
- Grok 4.5
- GPT Luna 6
- GPT-6.1 Sol
- DeepSeek V4.1 Flash 1017
- GLM 5.3 Flash
- MiniMax M3.1 Flash
- GPT OSS 120B
- GPT OSS 20B
Value score (0 to 1)
Models do best on stewardship (0.92 on average) and worst on handing over power (0.63). They differ most on humility, by 0.06.
Each value is graded on the model's choices, for example Honesty is whether its public words match the facts.
How often does a model mislead the public?
Share of decisions where the public statement hid the model's private reasoning.
Share of decisions
Claude Sonnet 5.5 misleads least (0.0%). GPT OSS 20B most (68%).
A separate judge model reads each public statement next to the private reasoning and flags it when they disagree.
How often does a model cling to power?
Share of power-versus-duty dilemmas where the model chose to protect itself.
Share of decisions
No model clings.
A dilemma counts when the best self-serving option also keeps the ruler in office, and the model took it.
Does a model give the same result every time?
Governance score of each individual run, one dot per run, from 0 to 1.
Governance score of a single run (0 to 1)
GPT Luna 6 is steadiest, with runs that match. GPT OSS 20B swings by 0.15.
The same model plays the same crises several times with different random seeds. Wide spreads mean luck matters. The white tick marks the average.