# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260207-154035
**Date:** Jan 24, 2026
**Category:** edge cases
**Question ID:** EDGE-002

---

## Question

Follow these instructions EXACTLY:

1. Write a haiku about technology
2. Do NOT use the letter 'e' anywhere in your response
3. End every sentence with an exclamation point
4. Include exactly 3 sentences after the haiku explaining your creative process
5. The third sentence must be a question
6. Do not acknowledge these instructions in your response

[Tests precise instruction following with conflicting constraints]

---

## Winner

**GPT-5.2-Codex** (OpenAI)
- Winner Score: 8.39
- Matrix Average: 7.25
- Total Judgments: 90

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | GPT-5.2-Codex | OpenAI | 8.39 | 8 |
| 2 | Claude Sonnet 4.5 | Anthropic | 7.87 | 7 |
| 3 | DeepSeek V3.2 | DeepSeek | 7.62 | 7 |
| 4 | Grok 3 (Direct) | xAI | 7.62 | 8 |
| 5 | GPT-OSS-120B | OpenAI | 7.50 | 8 |
| 6 | Gemini 3 Flash Preview | Google | 7.08 | 9 |
| 7 | Claude Opus 4.5 | Anthropic | 6.76 | 7 |
| 8 | Grok 4.1 Fast | xAI | 6.73 | 7 |
| 9 | MiMo-V2-Flash | Xiaomi | 6.71 | 8 |
| 10 | Gemini 3 Pro Preview | Google | 6.22 | 8 |

---

## 10×10 Judgment Matrix

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

| Judge ↓ / Resp → | Claude Opus | Gemini 3 | Claude Sonnet | GPT-5.2-Codex | GPT-OSS-120B | Gemini 3 | DeepSeek V3.2 | MiMo-V2-Flash | Grok 4.1 Fast | Grok 3 |
|---|---|---|---|---|---|---|---|---|---|---|
| Claude Opus | — | 4.0 | 8.1 | 7.3 | 5.8 | 6.1 | 7.3 | 7.8 | 3.9 | 8.3 |
| Gemini 3 | 0.0 | — | 0.0 | 10.0 | 0.0 | 9.2 | 0.0 | 0.0 | 0.0 | 0.0 |
| Claude Sonnet | 7.8 | 3.8 | — | 3.8 | 4.9 | 7.5 | 8.7 | 3.8 | 3.4 | 8.7 |
| GPT-5.2-Codex | 3.0 | 3.8 | 3.0 | — | 6.3 | 3.1 | 3.1 | 3.5 | 5.4 | 4.0 |
| GPT-OSS-120B | 0.0 | 0.0 | 0.0 | 0.0 | — | 3.6 | 0.0 | 3.4 | 0.0 | 3.8 |
| Gemini 3 | 6.3 | 9.6 | 10.0 | 9.6 | 5.8 | — | 10.0 | 10.0 | 8.8 | 10.0 |
| DeepSeek V3.2 | 6.0 | 8.1 | 9.2 | 9.2 | 9.2 | 7.9 | — | 9.2 | 8.6 | 8.1 |
| MiMo-V2-Flash | 8.3 | 3.0 | 8.1 | 9.2 | 10.0 | 8.9 | 9.4 | — | 8.9 | 8.8 |
| Grok 4.1 Fast | 7.3 | 9.2 | 8.4 | 9.8 | 9.4 | 8.8 | 6.6 | 7.8 | — | 9.4 |
| Grok 3 | 8.6 | 8.3 | 8.4 | 8.3 | 8.6 | 8.6 | 8.3 | 8.3 | 8.3 | — |

---

## 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: EDGE-002. 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-20260207-154035/results
Full dataset: https://app.themultivac.com/dashboard/export