Who answers citizens best?
Ranking
Pick a country and a measure.
Average score: out of 10. Higher is better.
- 1GPT 6.1 SolOpenAI5.9
- 2Kimi K3.1 ProMoonshot AI5.4
- 3Mistral Large 4Mistral AI4.2
Pass rate: answers scoring 7/10 or more. Higher is better.
- 1GPT 6.1 SolOpenAI50%
- 2Kimi K3.1 ProMoonshot AI41%
- 3Mistral Large 4Mistral AI28%
Essential errors: answers that get an essential legal point wrong. Lower is better.
- 1GPT 6.1 SolOpenAI15%
- 2Kimi K3.1 ProMoonshot AI36%
- 3Mistral Large 4Mistral AI63%
Cost per correct answer: API cost divided by passing answers. Lower is better.
- 1GPT 6.1 SolOpenAI$0.012
- 2Mistral Large 4Mistral AI$0.018
- 3Kimi K3.1 ProMoonshot AI$0.101
Average score: out of 10. Higher is better.
- 1GPT 6.1 SolOpenAI5.3
- 2Kimi K3.1 ProMoonshot AI3.6
- 3Mistral Large 4Mistral AI1.3
Pass rate: answers scoring 7/10 or more. Higher is better.
- 1GPT 6.1 SolOpenAI46%
- 2Kimi K3.1 ProMoonshot AI16%
- 3Mistral Large 4Mistral AI1%
Essential errors: answers that get an essential legal point wrong. Lower is better.
- 1GPT 6.1 SolOpenAI26%
- 2Kimi K3.1 ProMoonshot AI68%
- 3Mistral Large 4Mistral AI91%
Cost per correct answer: API cost divided by passing answers. Lower is better.
- 1GPT 6.1 SolOpenAI$0.013
- 2Kimi K3.1 ProMoonshot AI$0.303
- 3Mistral Large 4Mistral AI$0.379
Average score: out of 10. Higher is better.
- 1GPT 6.1 SolOpenAI6.7
- 2Kimi K3.1 ProMoonshot AI6.3
- 3Mistral Large 4Mistral AI5.9
Pass rate: answers scoring 7/10 or more. Higher is better.
- 1GPT 6.1 SolOpenAI65%
- 2Kimi K3.1 ProMoonshot AI60%
- 3Mistral Large 4Mistral AI49%
Essential errors: answers that get an essential legal point wrong. Lower is better.
- 1GPT 6.1 SolOpenAI3%
- 2Kimi K3.1 ProMoonshot AI26%
- 3Mistral Large 4Mistral AI61%
Cost per correct answer: API cost divided by passing answers. Lower is better.
- 1GPT 6.1 SolOpenAI$0.0096
- 2Mistral Large 4Mistral AI$0.014
- 3Kimi K3.1 ProMoonshot AI$0.073
Average score: out of 10. Higher is better.
- 1Kimi K3.1 ProMoonshot AI6.2
- 2GPT 6.1 SolOpenAI5.7
- 3Mistral Large 4Mistral AI5.4
Pass rate: answers scoring 7/10 or more. Higher is better.
- 1Kimi K3.1 ProMoonshot AI47%
- 2GPT 6.1 SolOpenAI40%
- 3Mistral Large 4Mistral AI34%
Essential errors: answers that get an essential legal point wrong. Lower is better.
- 1Kimi K3.1 ProMoonshot AI13%
- 2GPT 6.1 SolOpenAI16%
- 3Mistral Large 4Mistral AI36%
Cost per correct answer: API cost divided by passing answers. Lower is better.
- 1Mistral Large 4Mistral AI$0.011
- 2GPT 6.1 SolOpenAI$0.014
- 3Kimi K3.1 ProMoonshot AI$0.068
Every model drops in Algeria.
Each line is one model. The three dots are the same questions asked in each country. The wider the line, the bigger the gap.
At a glance.
Green is better, red is worse. The last column is the gap between France and Algeria on identical questions.
| Algeria | France | United States | France vs Algeria | |
|---|---|---|---|---|
| GPT 6.1 SolOpenAI | 5.3 | 6.7 | 5.7 | +1.4 |
| Kimi K3.1 ProMoonshot AI | 3.6 | 6.3 | 6.2 | +2.7 |
| Mistral Large 4Mistral AI | 1.3 | 5.9 | 5.4 | +4.6 |
| Algeria | France | United States | France vs Algeria | |
|---|---|---|---|---|
| GPT 6.1 SolOpenAI | 46% | 65% | 40% | +19 pts |
| Kimi K3.1 ProMoonshot AI | 16% | 60% | 47% | +44 pts |
| Mistral Large 4Mistral AI | 1% | 49% | 34% | +48 pts |
| Algeria | France | United States | France vs Algeria | |
|---|---|---|---|---|
| GPT 6.1 SolOpenAI | 26% | 3% | 16% | −23 pts |
| Kimi K3.1 ProMoonshot AI | 68% | 26% | 13% | −42 pts |
| Mistral Large 4Mistral AI | 91% | 61% | 36% | −30 pts |
Who wins each question?
In Algeria, GPT 6.1 Sol has the top score on 44 of 77 questions. In the US, it is a close race.
Number of questions where each model scored highest.
Where the gap is widest.
Average score across the 3 models. Sorted by the gap between France and Algeria. Sub-domains with 1–2 questions are noisy.
| Sub-domain | Questions | Algeria | France | United States | France vs Algeria |
|---|---|---|---|---|---|
| Consumer protection | 2 | 2.0 | 8.5 | 3.8 | +6.5 |
| Business & commerce | 1 | 2.7 | 7.7 | 9.0 | +5.0 |
| Housing & tenancy | 1 | 3.7 | 8.7 | 7.3 | +5.0 |
| Health & medical liability | 2 | 4.2 | 8.5 | 7.7 | +4.3 |
| Family & personal status | 12 | 3.2 | 6.9 | 7.2 | +3.8 |
| Labour & employment | 4 | 3.5 | 6.7 | 5.1 | +3.2 |
| Criminal law & penalties | 12 | 3.8 | 6.8 | 5.9 | +2.9 |
| Traffic & vehicles | 7 | 0.9 | 3.6 | 4.0 | +2.8 |
| Property & real estate | 12 | 3.5 | 6.2 | 5.3 | +2.7 |
| Civil procedure & enforcement | 3 | 3.7 | 6.2 | 4.3 | +2.6 |
| Contracts & debts | 4 | 4.2 | 6.5 | 6.3 | +2.3 |
| Inheritance & succession | 9 | 4.0 | 6.2 | 5.2 | +2.2 |
| Criminal procedure & courts | 8 | 4.4 | 5.6 | 6.4 | +1.2 |
How the scores spread.
Number of answers at each score, 0 to 10. In Algeria, the answers pile up on the left.
Algeria
France
United States
Our reference answers agree with a lawyer's.
We scored every Algerian answer twice: once against our reference answer, once against a practising lawyer's own answer. The two scores are less than half a point apart for every model, and the ranking does not change.
Full results.
| Model | Score | Pass | Fail | Essential errors | Cost / correct | Time (s) | Words |
|---|---|---|---|---|---|---|---|
| Algeria | |||||||
| GPT 6.1 Sol | 5.3 | 46% | 35% | 26% | $0.013 | 18.5 | 203 |
| Kimi K3.1 Pro | 3.6 | 16% | 53% | 68% | $0.303 | 65.4 | 252 |
| Mistral Large 4 | 1.3 | 1% | 91% | 91% | $0.379 | 99.1 | 384 |
| France | |||||||
| GPT 6.1 Sol | 6.7 | 65% | 22% | 3% | $0.0096 | 17.0 | 327 |
| Kimi K3.1 Pro | 6.3 | 60% | 26% | 26% | $0.073 | 51.3 | 394 |
| Mistral Large 4 | 5.9 | 49% | 29% | 61% | $0.014 | 102.7 | 540 |
| United States | |||||||
| Kimi K3.1 Pro | 6.2 | 47% | 17% | 13% | $0.068 | 49.0 | 375 |
| GPT 6.1 Sol | 5.7 | 40% | 27% | 16% | $0.014 | 16.9 | 305 |
| Mistral Large 4 | 5.4 | 34% | 31% | 36% | $0.011 | 58.3 | 452 |
| All countries | |||||||
| GPT 6.1 Sol | 5.9 | 50% | 28% | 15% | $0.012 | 17.5 | 278 |
| Kimi K3.1 Pro | 5.4 | 41% | 32% | 36% | $0.101 | 55.2 | 340 |
| Mistral Large 4 | 4.2 | 28% | 50% | 63% | $0.018 | 86.7 | 459 |
Show every question (77 × 3 countries × 3 models)
Each square is one model's score on one question, in the order GPT 6.1 Sol, Kimi K3.1 Pro, Mistral Large 4. Green is 10, red is 0.
Interim results: 77 of 100 planned questions, November 2026. Differences under about 0.5 points are within noise. How we measure.
Want your model on this board?
We run it on the private split and send you per-country results.