# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260402-181526
**Date:** Apr 02, 2026
**Category:** reasoning
**Question ID:** REASON-027

---

## Question

If you assume your birth rank among all humans who will ever live is randomly selected, and 100 billion humans have lived so far, then you should predict total humans ≈ 200 billion (50% chance you're in the second half). This implies humanity ends within a few centuries. (1) Is this reasoning valid? (2) What assumptions does it require? (3) Nick Bostrom's Self-Sampling Assumption vs Self-Indication Assumption — which is more reasonable and why?

---

## Winner

**Claude Sonnet 4.6** (openrouter)
- Winner Score: 9.12
- Matrix Average: 7.92
- Total Judgments: 80

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | Claude Sonnet 4.6 | openrouter | 9.13 | 8 |
| 2 | Claude Opus 4.6 | openrouter | 9.05 | 9 |
| 3 | GPT-5.4 | openrouter | 8.96 | 9 |
| 4 | Grok 4.20 | openrouter | 8.74 | 9 |
| 5 | Gemini 2.5 Flash | openrouter | 8.47 | 9 |
| 6 | DeepSeek V4 | openrouter | 8.08 | 9 |
| 7 | MiniMax M2.5 | openrouter | 7.96 | 9 |
| 8 | Gemini 3.1 Pro | openrouter | 7.38 | 9 |
| 9 | GPT-OSS-120B | OpenAI | 6.15 | 5 |
| 10 | MiMo-V2-Flash | Xiaomi | 5.26 | 4 |

---

## 10×10 Judgment Matrix

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

| Judge ↓ / Resp → | DeepSeek V4 | MiMo-V2-Flash | Gemini 3.1 Pro | Claude Opus | GPT-5.4 | Grok 4.20 | Claude Sonnet | GPT-OSS-120B | Gemini 2.5 | MiniMax M2.5 |
|---|---|---|---|---|---|---|---|---|---|---|
| DeepSeek V4 | — | 8.1 | 8.7 | 8.8 | 8.8 | 8.8 | 9.4 | 7.5 | 8.8 | 8.8 |
| MiMo-V2-Flash | 8.4 | — | 8.4 | 9.4 | 9.1 | 9.0 | 9.3 | 8.0 | 8.7 | 9.1 |
| Gemini 3.1 Pro | 7.0 | · | — | 9.8 | 9.7 | 10.0 | 10.0 | · | 9.7 | 7.0 |
| Claude Opus | 8.0 | 0.5 | 7.3 | — | 8.8 | 9.2 | 9.6 | 0.5 | 7.6 | 7.8 |
| GPT-5.4 | 7.8 | · | 5.5 | 9.0 | — | 7.3 | 8.8 | · | 8.0 | 6.3 |
| Grok 4.20 | 8.4 | 5.0 | 8.4 | 8.8 | 8.8 | — | 8.8 | 7.4 | 8.7 | 7.8 |
| Claude Sonnet | 8.0 | · | 7.0 | 8.8 | 9.2 | 9.2 | — | · | 8.4 | 7.8 |
| GPT-OSS-120B | 8.0 | · | 5.7 | 8.8 | 8.4 | 7.7 | · | — | 8.4 | 8.2 |
| Gemini 2.5 | 9.0 | 7.5 | 8.4 | 8.8 | 8.8 | 9.7 | 8.8 | 7.5 | — | 8.8 |
| MiniMax M2.5 | 8.2 | · | 6.9 | 9.0 | 8.8 | 7.8 | 8.2 | · | 8.0 | — |

---

## 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-027. 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-20260402-181526/results
Full dataset: https://app.themultivac.com/dashboard/export