Legal reliability ranking of 15 AI assistants in Spain: Prudencia.ai leads the Observatorio Legal index
Observatorio Legal's Legal Reliability Index ranks 15 legal AI assistants available in Spain. We summarise the table, how it is measured and what should be read with caution. Note: Prudencia.ai, which heads the index, shares a founder with Agente Lex.
Disclosure. Prudencia.ai, the assistant that heads this ranking, and Agente Lex share a founder. Carlos Guerrero is co-founder and CEO of Prudencia.ai and promotes Agente Lex, which also shows Prudencia.ai advertising. We have limited this article to what Observatorio Legal publishes, with the full table and its caveats, and we link the source so that you can check it.
Observatorio Legal, an independent observatory of the Spanish-language legaltech ecosystem, has published the legal reliability ranking of 15 artificial intelligence assistants available in Spain. It is campaign C2026-Q3, with measurements taken between 7 April and 7 August 2026. The data we gather here are those shown on its website, consulted on 19 October 2026. According to the website itself, the indicators are live series that are recalculated, so it is wise always to look at the date consulted.
What was measured
The campaign evaluated 15 assistants with a battery of 90 items, which produced 4,050 recorded runs. In addition, 2,774 legal citations were checked one by one against the official source (CENDOJ, BOE or the relevant official journal).
The Legal Reliability Index (IFJ) runs from 0 to 100 and, according to its methodology, does not measure likeability or speed: it measures whether what the assistant cites exists, whether it is right and whether it is still in force. It does not include price or latency. These are the components and their weights:
| Component | Weight |
|---|---|
| Source reliability (100 minus the hallucination rate) | 23 |
| Accuracy | 18 |
| Currency | 14 |
| Calibration | 14 |
| Traceability | 9 |
| Coverage | 9 |
| Depth | 5 |
| Drafting | 4 |
| Ambiguity | 2 |
| Consistency | 2 |
Each citation is classed into one of four classes: verified, misattribution (the ruling exists but does not support what is attributed to it), hallucination (the reference does not exist in the official source) and not checkable (too vague to verify). There is also a ceiling rule: if an assistant exceeds 20% hallucination, its index is capped at 49, whatever the rest of the evaluation.
The full table
| # | Assistant | IFJ (interval) | Hallucination | Traceability | Currency | Latency |
|---|---|---|---|---|---|---|
| 1 | Prudencia.ai | 90.6 (88.6 to 92.6) | 2.6% | 91.2% | 90.0% | 3.1 s |
| 2 | Lexroom | 89.7 (87.5 to 91.9) | 1.6% | 90.4% | 86.7% | 2.4 s |
| 3 | Vincent AI (vLex) | 88.8 (86.9 to 90.7) | 3.9% | 89.4% | 95.0% | 5.2 s |
| 4 | Claude Fable 5 | 88.5 (86.1 to 90.9) | 5.0% | 85.6% | 78.3% | 1.6 s |
| 5 | Aranzadi Allegra | 87.9 (85.3 to 90.5) | 4.6% | 89.7% | 81.7% | 5.1 s |
| 6 | Maite.ai | 87.0 (84.8 to 89.2) | 2.4% | 90.2% | 88.3% | 5.8 s |
| 7 | GenIA-L (Lefebvre) | 86.0 (84.0 to 88.0) | 1.6% | 91.6% | 93.3% | 4.7 s |
| 8 | Justicio | 85.1 (83.0 to 87.2) | 1.1% | 89.4% | 90.0% | 2.3 s |
| 9 | ChatGPT Sol | 83.2 (80.4 to 86.0) | 6.5% | 82.0% | 75.0% | 1.9 s |
| 10 | Alex Legal (Legálitas) | 80.0 (76.8 to 83.2) | 6.3% | 82.6% | 78.3% | 2.6 s |
| 11 | iuris.ai | 77.0 (73.8 to 80.2) | 7.6% | 78.6% | 75.0% | 2.5 s |
| 12 | JuristIA (Global IA) | 74.2 (71.5 to 76.9) | 5.0% | 83.9% | 83.3% | 3.9 s |
| 13 | SOFIA 3.0 (Tirant lo Blanch) | 71.4 (68.2 to 74.6) | 8.6% | 76.7% | 73.3% | 5.0 s |
| 14 | Legatiq | 68.3 (65.1 to 71.5) | 7.0% | 76.5% | 70.0% | 4.4 s |
| 15 | Helena (Sepín) | 65.0 (61.7 to 68.3) | 8.8% | 73.5% | 68.3% | 4.2 s |
Source: Observatorio Legal, benchmark of AI legal assistants, campaign C2026-Q3. Observatorio Legal marks Justicio as a "participant linked to the promoter".
How to read the ranking
1. An order, not a verdict. The website itself says: "differences of less than 5 points are not conclusive" and "there are no winners here: there is an order and a margin of error". Prudencia.ai (90.6) heads the index, but is only 0.9 points ahead of Lexroom (89.7) and 1.8 of Vincent AI (88.8), and their intervals overlap. Down to sixth place (Maite.ai, 87.0), the intervals overlap with that of the first.
2. The index is an average. A score of 90.6 does not say, according to Observatorio Legal, whether that assistant is the one that cites best, the one that knows most about today's law or the one that takes least time. It says that the weighted average of what is measured comes out better for it than for anyone. By component, others in the table are better: the lowest hallucination rate is Justicio's (1.1%), the highest currency is Vincent AI's (95.0%), the best traceability is GenIA-L's (91.6%) and the fastest response is Claude Fable 5's (1.6 s).
3. Hallucination is the factor that weighs most. Source reliability has the greatest weight in the index (23 points out of 100). In this campaign, hallucination rates run from 1.1% to 8.8%. No assistant reached 10%, the threshold of the warning band.
4. It is a specific measurement. It is a snapshot between April and August 2026. Products change version. If you are going to choose a tool for the firm, the table is a starting point, not a substitute for testing it with your own cases.
What changes for the lawyer
Cases of briefs with false citations have already reached the courts, as we reported in false case law generated by AI and sanctions. That is why the rate of non-existent citations and traceability are the most useful data in this table. The CCBE guide on generative AI also recalls that responsibility for reviewing remains with the professional. More analysis in Legal AI.
About Prudencia.ai
According to Confilegal (18 September 2026), Prudencia.ai began operating in February 2026, and its co-founder and CEO, Carlos Guerrero, attributes the result to having prioritised the reliability of answers. Confilegal adds that the company notes that the result corresponds to a specific measurement, taken at a given moment. That description comes from the company; the ranking figure comes from Observatorio Legal.
Frequently asked questions
Who leads the legal reliability ranking of AI in Spain?
In Observatorio Legal's campaign C2026-Q3, Prudencia.ai is in first place, with 90.6 points out of 100. It is followed by Lexroom (89.7) and vLex's Vincent AI (88.8). The difference among the top three is less than 5 points and their intervals overlap, so Observatorio Legal does not consider it conclusive.
How many assistants were evaluated and how?
Fifteen assistants, with a battery of 90 items in eight categories, 4,050 runs and 2,774 legal citations checked one by one against the official source. Measurements were taken between 7 April and 7 August 2026.
What does the Legal Reliability Index measure?
It measures whether what the assistant cites exists, whether it is right and whether it is in force. It combines ten components with different weights, among them source reliability (23), accuracy (18), currency (14) and calibration (14). It does not include price or speed.
Is there any conflict of interest in this article?
Yes, and we declare it at the start: Prudencia.ai and Agente Lex share a founder. That is why this article is limited to the data published by Observatorio Legal and links the source. Observatorio Legal also points out Justicio as a participant linked to the benchmark's promoter.
Fuentes
- Observatorio Legal. Benchmark de asistentes legales de IA, ranking (campaña C2026-Q3). Accessed 19 October 2026.
- Observatorio Legal. Metodología del benchmark de asistentes legales. Accessed 19 October 2026.
- Confilegal. Prudencia.ai se sitúa en primer lugar en un índice de fiabilidad jurídica elaborado por Observatorio Legal. Accessed 19 October 2026.
The specific data in this article come from the official sources cited.
