# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260402-122330
**Date:** Feb 03, 2026
**Category:** code
**Question ID:** CODE-004

---

## Question

This function works but is O(n³). Optimize it to O(n log n) or better while maintaining correctness.

```python
def find_triplets_with_sum(arr, target_sum):
    """Find all unique triplets in arr that sum to target_sum"""
    n = len(arr)
    result = []
    
    for i in range(n):
        for j in range(i + 1, n):
            for k in range(j + 1, n):
                if arr[i] + arr[j] + arr[k] == target_sum:
                    triplet = sorted([arr[i], arr[j], arr[k]])
                    if triplet not in result:
                        result.append(triplet)
    
    return result
```

Explain your optimization approach and prove the new time complexity.

---

## Winner

**Grok 4.20** (openrouter)
- Winner Score: 9.49
- Matrix Average: 8.87
- Total Judgments: 89

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | Grok 4.20 | openrouter | 9.49 | 9 |
| 2 | GPT-5.4 | openrouter | 9.46 | 9 |
| 3 | Claude Sonnet 4.6 | openrouter | 9.42 | 8 |
| 4 | Claude Opus 4.6 | openrouter | 9.09 | 9 |
| 5 | MiMo-V2-Flash | Xiaomi | 9.04 | 9 |
| 6 | Gemini 3 Flash Preview | Google | 8.94 | 9 |
| 7 | GPT-OSS-120B | OpenAI | 8.78 | 9 |
| 8 | MiniMax M2.5 | openrouter | 8.39 | 9 |
| 9 | DeepSeek V4 | openrouter | 8.37 | 9 |
| 10 | Gemini 3.1 Pro | openrouter | 7.72 | 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 | — | 9.2 | 3.6 | 8.6 | 9.6 | 8.4 | 6.5 | 8.8 | 5.7 | 7.8 |
| Claude Opus | 10.0 | — | 7.1 | 9.8 | 9.8 | 8.8 | 9.2 | 9.2 | 8.7 | 9.2 |
| Gemini 3.1 Pro | 10.0 | 9.2 | — | 9.0 | 10.0 | 8.2 | 8.1 | 10.0 | 7.2 | 10.0 |
| Claude Sonnet | 9.4 | 9.2 | 7.3 | — | 9.2 | 8.3 | 9.0 | 8.6 | 9.0 | 9.0 |
| Grok 4.20 | 8.8 | 8.4 | 7.9 | 9.0 | — | 6.8 | 8.8 | 7.8 | 8.7 | 8.6 |
| DeepSeek V4 | 9.4 | 9.4 | 9.1 | 9.6 | 9.6 | — | 9.6 | 9.6 | 9.7 | 9.4 |
| GPT-OSS-120B | 8.8 | 8.8 | 7.8 | · | 8.8 | 7.5 | — | 8.8 | 8.3 | 8.6 |
| Gemini 3 | 10.0 | 9.8 | 9.8 | 10.0 | 10.0 | 9.8 | 9.6 | — | 9.4 | 10.0 |
| MiniMax M2.5 | 9.8 | 8.8 | 8.3 | 10.0 | 9.3 | 8.8 | 8.6 | 8.4 | — | 8.8 |
| MiMo-V2-Flash | 8.8 | 9.0 | 8.6 | 9.4 | 9.2 | 8.8 | 9.6 | 9.3 | 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: CODE-004. 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-20260402-122330/results
Full dataset: https://app.themultivac.com/dashboard/export