# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260402-233502
**Date:** Apr 02, 2026
**Category:** communication
**Question ID:** COMM-021

---

## Question

Write a 60-second elevator pitch for each of these: (1) A startup that uses AI to detect infrastructure defects from drone footage. Audience: a VC partner at a cocktail party. (2) The same startup. Audience: a city infrastructure manager at a conference. (3) The same startup. Audience: your grandmother at Thanksgiving dinner. Each should be natural, conversational, and end with a clear next step.

---

## Winner

**Claude Opus 4.6** (openrouter)
- Winner Score: 9.22
- Matrix Average: 8.84
- Total Judgments: 89

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | Claude Opus 4.6 | openrouter | 9.22 | 9 |
| 2 | Mistral Small Creative | Mistral | 9.16 | 8 |
| 3 | Grok 4.20 | openrouter | 9.09 | 9 |
| 4 | GPT-OSS-120B | OpenAI | 9.08 | 9 |
| 5 | Claude Sonnet 4.6 | openrouter | 9.04 | 9 |
| 6 | MiMo-V2-Flash | Xiaomi | 8.99 | 9 |
| 7 | GPT-5.4 | openrouter | 8.98 | 9 |
| 8 | Seed 1.6 Flash | openrouter | 8.85 | 9 |
| 9 | DeepSeek V4 | openrouter | 8.64 | 9 |
| 10 | Gemini 3.1 Pro | openrouter | 7.33 | 9 |

---

## 10×10 Judgment Matrix

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

| Judge ↓ / Resp → | Claude Opus | GPT-5.4 | Claude Sonnet | Gemini 3.1 Pro | Grok 4.20 | DeepSeek V4 | GPT-OSS-120B | MiMo-V2-Flash | Mistral Small | Seed 1.6 Flash |
|---|---|---|---|---|---|---|---|---|---|---|
| Claude Opus | — | 9.0 | 9.6 | 8.1 | 9.0 | 8.3 | 9.0 | 9.3 | 8.6 | 8.6 |
| GPT-5.4 | 9.0 | — | 8.4 | 4.0 | 8.8 | 8.3 | 8.6 | 8.2 | 9.2 | 7.8 |
| Claude Sonnet | 9.6 | 9.2 | — | 8.0 | 9.2 | 8.3 | 9.0 | 9.2 | · | 8.6 |
| Gemini 3.1 Pro | 9.8 | 9.8 | 10.0 | — | 10.0 | 9.6 | 10.0 | 10.0 | 9.8 | 10.0 |
| Grok 4.20 | 8.8 | 9.0 | 8.3 | 5.8 | — | 8.1 | 8.8 | 8.3 | 9.0 | 8.3 |
| DeepSeek V4 | 9.3 | 8.8 | 8.6 | 8.6 | 9.3 | — | 9.0 | 9.0 | 9.8 | 9.0 |
| GPT-OSS-120B | 8.7 | 8.3 | 8.7 | 5.3 | 8.3 | 8.2 | — | 8.3 | 8.3 | 8.4 |
| MiMo-V2-Flash | 9.6 | 9.0 | 9.6 | 8.6 | 9.0 | 9.0 | 9.0 | — | 9.6 | 9.6 |
| Mistral Small | 9.6 | 9.6 | 9.6 | 9.6 | 9.6 | 9.4 | 9.6 | 9.8 | — | 9.6 |
| Seed 1.6 Flash | 8.8 | 8.2 | 8.8 | 8.0 | 8.8 | 8.4 | 8.8 | 8.8 | 9.0 | — |

---

## 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: COMM-021. 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-20260402-233502/results
Full dataset: https://app.themultivac.com/dashboard/export