# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260402-140852
**Date:** Apr 02, 2026
**Category:** code
**Question ID:** CODE-021

---

## Question

Write a Python function that parses unified diff format (the output of `git diff`) and returns a structured representation: files changed, lines added/removed/modified, hunks with context. Handle edge cases: binary files, renamed files, mode changes, and empty diffs. Include tests with real diff examples.

---

## Winner

**Gemini 3 Flash Preview** (Google)
- Winner Score: 8.03
- Matrix Average: 6.01
- Total Judgments: 74

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | Gemini 3 Flash Preview | Google | 8.03 | 9 |
| 2 | GPT-5.4 | openrouter | 7.92 | 8 |
| 3 | Grok 4.20 | openrouter | 7.39 | 8 |
| 4 | Claude Opus 4.6 | openrouter | 6.53 | 9 |
| 5 | DeepSeek V4 | openrouter | 6.46 | 8 |
| 6 | GPT-OSS-120B | OpenAI | 6.28 | 8 |
| 7 | Claude Sonnet 4.6 | openrouter | 5.49 | 8 |
| 8 | MiMo-V2-Flash | Xiaomi | 4.36 | 9 |
| 9 | Gemini 3.1 Pro | openrouter | 1.62 | 7 |

---

## 10×10 Judgment Matrix

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

| Judge ↓ / Resp → | GPT-5.4 | Claude Opus | Gemini 3.1 Pro | Claude Sonnet | Gemini 3 | Grok 4.20 | DeepSeek V4 | GPT-OSS-120B | MiniMax M2.5 | MiMo-V2-Flash |
|---|---|---|---|---|---|---|---|---|---|---|
| GPT-5.4 | — | 2.3 | 0.7 | 1.6 | 6.8 | 6.0 | 5.0 | 2.8 | · | 2.5 |
| Claude Opus | 7.8 | — | 0.7 | 4.5 | 7.8 | 6.4 | 5.4 | 5.3 | · | 3.3 |
| Gemini 3.1 Pro | 7.4 | 4.9 | — | 4.7 | 9.0 | 7.3 | · | 4.5 | · | 3.3 |
| Claude Sonnet | 8.0 | 6.3 | 0.2 | — | 7.3 | 7.2 | 6.4 | 7.0 | · | 4.0 |
| Gemini 3 | 9.0 | 8.4 | 1.6 | 7.0 | — | 9.2 | 8.2 | 7.0 | · | 5.8 |
| Grok 4.20 | 7.0 | 7.9 | · | · | 7.8 | — | 6.2 | · | · | 4.4 |
| DeepSeek V4 | 8.6 | 9.0 | 4.0 | 8.6 | 8.6 | · | — | 8.8 | · | 8.6 |
| GPT-OSS-120B | 7.8 | 4.8 | 1.4 | 4.2 | 8.3 | 7.0 | 6.2 | — | · | 2.9 |
| MiniMax M2.5 | · | 6.5 | · | 6.3 | 8.0 | 7.5 | 7.3 | 6.7 | — | 4.5 |
| MiMo-V2-Flash | 7.8 | 8.6 | 2.8 | 7.0 | 8.6 | 8.6 | 7.2 | 8.1 | · | — |

---

## 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-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-20260402-140852/results
Full dataset: https://app.themultivac.com/dashboard/export