# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260402-145527
**Date:** Apr 02, 2026
**Category:** code
**Question ID:** CODE-027

---

## Question

Implement a JSON Schema validator from scratch that supports: type validation (string, number, integer, boolean, array, object, null), required fields, minimum/maximum for numbers, minLength/maxLength for strings, pattern (regex), enum, nested objects, and array items validation. No external libraries.

---

## Winner

**GPT-5.4** (openrouter)
- Winner Score: 8.96
- Matrix Average: 6.77
- Total Judgments: 90

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | GPT-5.4 | openrouter | 8.96 | 9 |
| 2 | DeepSeek V4 | openrouter | 8.66 | 9 |
| 3 | Grok 4.20 | openrouter | 8.61 | 9 |
| 4 | Gemini 3 Flash Preview | Google | 8.20 | 9 |
| 5 | GPT-OSS-120B | OpenAI | 7.55 | 9 |
| 6 | MiMo-V2-Flash | Xiaomi | 7.35 | 9 |
| 7 | Claude Sonnet 4.6 | openrouter | 6.22 | 9 |
| 8 | Claude Opus 4.6 | openrouter | 5.62 | 9 |
| 9 | MiniMax M2.5 | openrouter | 3.59 | 9 |
| 10 | Gemini 3.1 Pro | openrouter | 2.92 | 9 |

---

## 10×10 Judgment Matrix

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

| Judge ↓ / Resp → | GPT-OSS-120B | GPT-5.4 | Claude Opus | Gemini 3.1 Pro | Claude Sonnet | Grok 4.20 | DeepSeek V4 | Gemini 3 | MiniMax M2.5 | MiMo-V2-Flash |
|---|---|---|---|---|---|---|---|---|---|---|
| GPT-OSS-120B | — | 8.8 | 3.8 | 2.1 | 4.4 | 8.8 | 9.0 | 8.8 | 3.4 | 8.3 |
| GPT-5.4 | 4.4 | — | 2.5 | 0.7 | 3.5 | 7.8 | 8.0 | 8.3 | 1.4 | 5.2 |
| Claude Opus | 7.4 | 9.0 | — | 1.6 | 5.3 | 7.8 | 8.6 | 8.0 | 1.9 | 7.0 |
| Gemini 3.1 Pro | 5.2 | 9.8 | 4.5 | — | 6.4 | 8.2 | 9.4 | 7.8 | 3.2 | 6.3 |
| Claude Sonnet | 7.8 | 8.6 | 6.8 | 1.9 | — | 8.8 | 8.6 | 7.8 | 4.3 | 7.4 |
| Grok 4.20 | 8.6 | 8.6 | 8.5 | 0.8 | 5.8 | — | 7.3 | 7.0 | 5.5 | 6.8 |
| DeepSeek V4 | 9.4 | 8.6 | 8.8 | 8.2 | 8.4 | 8.6 | — | 8.8 | 7.0 | 8.6 |
| Gemini 3 | 9.2 | 9.8 | 6.3 | 2.3 | 7.8 | 9.8 | 9.8 | — | 5.5 | 9.6 |
| MiniMax M2.5 | 7.4 | 8.8 | 5.0 | 0.4 | 6.7 | 8.8 | 8.6 | 8.6 | — | 7.0 |
| MiMo-V2-Flash | 8.6 | 8.6 | 4.4 | 8.3 | 7.7 | 9.0 | 8.8 | 8.6 | 0.2 | — |

---

## 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-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-20260402-145527/results
Full dataset: https://app.themultivac.com/dashboard/export