# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260403-125023
**Date:** Apr 03, 2026
**Category:** code
**Question ID:** CODE-016

---

## Question

Given these hex dumps of network packets and their known meanings, reverse-engineer the binary protocol format and write a parser.

Packet 1 (Login): 4d 56 01 00 0c 68 65 6c 6c 6f 5f 77 6f 72 6c 64 00 00 00 05 61 64 6d 69 6e
Known: username="hello_world", password="admin"

Packet 2 (Login): 4d 56 01 00 08 74 65 73 74 75 73 65 72 00 00 00 04 70 61 73 73
Known: username="testuser", password="pass"

Packet 3 (Message): 4d 56 02 00 05 68 65 6c 6c 6f 00 00 00 01
Known: message="hello", room_id=1

Write the protocol specification and a Python parser/serializer.

---

## Winner

**GPT-OSS-120B** (OpenAI)
- Winner Score: 9.38
- Matrix Average: 6.19
- Total Judgments: 76

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | GPT-OSS-120B | OpenAI | 9.38 | 9 |
| 2 | GPT-5.4 | openrouter | 9.08 | 9 |
| 3 | Gemini 3 Flash Preview | Google | 7.37 | 9 |
| 4 | Grok 4.20 | openrouter | 7.29 | 7 |
| 5 | MiMo-V2-Flash | Xiaomi | 6.81 | 9 |
| 6 | DeepSeek V4 | openrouter | 5.46 | 8 |
| 7 | Claude Sonnet 4.6 | openrouter | 4.62 | 9 |
| 8 | Claude Opus 4.6 | openrouter | 4.32 | 7 |
| 9 | Gemini 3.1 Pro | openrouter | 1.35 | 9 |

---

## 10×10 Judgment Matrix

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

| Judge ↓ / Resp → | GPT-5.4 | Claude Opus | Gemini 3.1 Pro | Claude Sonnet | Grok 4.20 | DeepSeek V4 | GPT-OSS-120B | Gemini 3 | MiniMax M2.5 | MiMo-V2-Flash |
|---|---|---|---|---|---|---|---|---|---|---|
| GPT-5.4 | — | 1.9 | 0.5 | 2.3 | 3.0 | 3.8 | 9.0 | 4.2 | · | 4.2 |
| Claude Opus | 8.9 | — | 0.7 | 4.5 | 5.2 | 5.0 | 9.6 | 6.2 | · | 5.3 |
| Gemini 3.1 Pro | 9.0 | 2.5 | — | 4.0 | 5.2 | 3.8 | 9.8 | 7.5 | · | 6.3 |
| Claude Sonnet | 8.8 | 4.3 | 1.0 | — | 8.8 | 5.3 | 9.6 | 9.0 | · | 6.3 |
| Grok 4.20 | 8.8 | · | 2.0 | 4.3 | — | 5.5 | 7.5 | 8.8 | · | 5.5 |
| DeepSeek V4 | 9.6 | 6.3 | 1.9 | 5.8 | 9.6 | — | 9.8 | 8.6 | · | 9.6 |
| GPT-OSS-120B | 8.8 | 4.3 | 0.7 | 2.9 | · | 4.3 | — | 4.6 | · | 4.3 |
| Gemini 3 | 10.0 | · | 1.6 | 6.8 | 10.0 | 7.3 | 10.0 | — | · | 10.0 |
| MiniMax M2.5 | 8.8 | 4.2 | 1.0 | 5.2 | · | · | 9.6 | 8.8 | — | 9.6 |
| MiMo-V2-Flash | 9.0 | 6.8 | 2.9 | 5.8 | 9.3 | 8.6 | 9.6 | 8.6 | · | — |

---

## 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: CODE-016. 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-20260403-125023/results
Full dataset: https://app.themultivac.com/dashboard/export