# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260402-154152
**Date:** Feb 04, 2026
**Category:** reasoning
**Question ID:** REASON-004

---

## Question

On an island, there are 100 people with blue eyes and 100 people with brown eyes. Everyone can see everyone else's eye color but not their own. There are no mirrors or ways to discover your own eye color.

There's a rule: if you ever figure out your own eye color, you must leave the island at midnight that day.

One day, a visitor announces to everyone: "At least one person on this island has blue eyes."

What happens? When? Why?

Be precise about the logical chain of events.

---

## Winner

**Grok 4.20** (openrouter)
- Winner Score: 9.66
- Matrix Average: 9.05
- Total Judgments: 81

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | Grok 4.20 | openrouter | 9.66 | 9 |
| 2 | Claude Opus 4.6 | openrouter | 9.60 | 9 |
| 3 | GPT-OSS-120B | OpenAI | 9.60 | 9 |
| 4 | GPT-5.4 | openrouter | 9.55 | 9 |
| 5 | Claude Sonnet 4.6 | openrouter | 9.48 | 9 |
| 6 | MiMo-V2-Flash | Xiaomi | 9.45 | 9 |
| 7 | DeepSeek V4 | openrouter | 9.43 | 9 |
| 8 | Gemini 3.1 Pro | openrouter | 7.57 | 9 |
| 9 | Gemini 2.5 Flash | openrouter | 7.10 | 9 |

---

## 10×10 Judgment Matrix

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

| Judge ↓ / Resp → | GPT-OSS-120B | Gemini 3.1 Pro | DeepSeek V4 | Claude Opus | GPT-5.4 | Grok 4.20 | Claude Sonnet | MiMo-V2-Flash | Gemini 2.5 | MiniMax M2.5 |
|---|---|---|---|---|---|---|---|---|---|---|
| GPT-OSS-120B | — | 5.8 | 9.1 | 8.7 | 8.9 | 9.1 | 8.4 | 8.7 | 4.8 | · |
| Gemini 3.1 Pro | 9.4 | — | 10.0 | 10.0 | 10.0 | 10.0 | 10.0 | 10.0 | 6.0 | · |
| DeepSeek V4 | 9.7 | 8.4 | — | 9.7 | 9.7 | 9.5 | 9.7 | 9.5 | 9.7 | · |
| Claude Opus | 9.6 | 7.3 | 9.4 | — | 9.4 | 9.7 | 9.7 | 10.0 | 7.5 | · |
| GPT-5.4 | 10.0 | 5.5 | 9.8 | 9.8 | — | 9.8 | 9.4 | 8.8 | 5.8 | · |
| Grok 4.20 | 8.8 | 7.9 | 8.7 | 8.8 | 8.8 | — | 8.7 | 8.8 | 7.5 | · |
| Claude Sonnet | 9.6 | 8.1 | 8.8 | 10.0 | 9.8 | 9.7 | — | 9.4 | 8.0 | · |
| MiMo-V2-Flash | 10.0 | 8.8 | 10.0 | 10.0 | 10.0 | 10.0 | 10.0 | — | 8.6 | · |
| Gemini 2.5 | 9.8 | 9.7 | 9.8 | 9.8 | 9.8 | 9.8 | 9.8 | 9.8 | — | · |
| MiniMax M2.5 | 9.4 | 6.7 | 9.3 | 9.6 | 9.4 | 9.4 | 9.6 | 10.0 | 6.0 | — |

---

## 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: REASON-004. 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-20260402-154152/results
Full dataset: https://app.themultivac.com/dashboard/export