# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260402-210327
**Date:** Apr 02, 2026
**Category:** analysis
**Question ID:** ANALYSIS-026

---

## Question

Country X is debating a points-based immigration system. Proposed criteria: education (30%), work experience (25%), age (20%), language proficiency (15%), job offer (10%). (1) Analyze potential biases in this system. (2) What outcomes would you measure to evaluate success after 5 years? (3) Country Y uses a lottery system instead. Compare the two approaches using economic, social, and ethical dimensions. (4) Design a hybrid system that addresses weaknesses of both.

---

## Winner

**Grok 4.20** (openrouter)
- Winner Score: 9.26
- Matrix Average: 8.49
- Total Judgments: 86

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | Grok 4.20 | openrouter | 9.26 | 9 |
| 2 | MiMo-V2-Flash | Xiaomi | 8.93 | 9 |
| 3 | GPT-OSS-120B | OpenAI | 8.86 | 9 |
| 4 | Gemini 3 Flash Preview | Google | 8.84 | 8 |
| 5 | MiniMax M2.5 | openrouter | 8.69 | 8 |
| 6 | DeepSeek V4 | openrouter | 8.59 | 9 |
| 7 | GPT-5.4 | openrouter | 8.43 | 9 |
| 8 | Claude Sonnet 4.6 | openrouter | 8.09 | 9 |
| 9 | Claude Opus 4.6 | openrouter | 7.97 | 7 |
| 10 | Gemini 3.1 Pro | openrouter | 7.26 | 9 |

---

## 10×10 Judgment Matrix

Rows = Judge, Columns = Respondent. Self-judgments excluded (—).

| Judge ↓ / Resp → | MiMo-V2-Flash | Gemini 3.1 Pro | Claude Opus | GPT-5.4 | DeepSeek V4 | Claude Sonnet | Grok 4.20 | GPT-OSS-120B | Gemini 3 | MiniMax M2.5 |
|---|---|---|---|---|---|---|---|---|---|---|
| MiMo-V2-Flash | — | 8.4 | 7.8 | 7.8 | 9.0 | 7.8 | 9.3 | 8.8 | 9.0 | 9.2 |
| Gemini 3.1 Pro | 9.8 | — | 7.5 | 7.3 | 9.7 | 7.9 | 10.0 | 9.1 | 10.0 | 9.7 |
| Claude Opus | 8.4 | 6.3 | — | 7.8 | 8.0 | 8.1 | 9.0 | 8.1 | · | · |
| GPT-5.4 | 8.4 | 5.8 | · | — | 8.4 | 6.8 | 9.2 | 8.2 | 8.8 | 8.2 |
| DeepSeek V4 | 9.2 | 8.4 | 8.8 | 8.8 | — | 8.7 | 9.0 | 9.0 | 8.8 | 9.0 |
| Claude Sonnet | 8.8 | 7.8 | 8.4 | 8.7 | 8.0 | — | 9.2 | 9.2 | 8.8 | 8.2 |
| Grok 4.20 | 8.7 | 7.5 | · | 8.7 | 8.4 | 8.8 | — | 8.7 | 8.7 | 8.3 |
| GPT-OSS-120B | 8.4 | 5.5 | 6.0 | 8.4 | 7.8 | 7.5 | 8.8 | — | 8.3 | 8.0 |
| Gemini 3 | 9.8 | 8.7 | 9.4 | 9.4 | 9.8 | 9.7 | 9.8 | 9.8 | — | 9.0 |
| MiniMax M2.5 | 8.8 | 6.8 | 7.8 | 8.8 | 8.2 | 7.7 | 9.0 | 9.0 | 8.4 | — |

---

## Methodology

- **10×10 Blind Peer Matrix:** All models answer the same question, then all models judge all responses.
- **5 Criteria:** Correctness, completeness, clarity, depth, usefulness (each scored 1–10).
- **Self-judgments excluded:** Models do not judge their own responses.
- **Weighted Score:** Composite of all 5 criteria.

---

## Citation

The Multivac (2026). Blind Peer Evaluation: ANALYSIS-026. app.themultivac.com

## License

Open data. Free to use, share, and build upon. Please cite The Multivac when using this data.

Download raw JSON: https://app.themultivac.com/api/evaluations/EVAL-20260402-210327/results
Full dataset: https://app.themultivac.com/dashboard/export