# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260402-123643
**Date:** Feb 24, 2026
**Category:** code
**Question ID:** CODE-007

---

## Question

Explain what this code does in plain English. Then identify any bugs or design issues.

```python
def f(x, n=3, m=None):
    m = m or {}
    if n == 0:
        return [[]]
    if x in m:
        return m[x]
    r = []
    for i in range(len(x)):
        for p in f(x[:i] + x[i+1:], n-1, m):
            r.append([x[i]] + p)
    m[x] = r
    return r

def g(s, k):
    from collections import Counter
    c = Counter(s)
    h = []
    import heapq
    for ch, cnt in c.items():
        heapq.heappush(h, (-cnt, ch))
    r = []
    while h and len(r) < k:
        cnt, ch = heapq.heappop(h)
        r.append(ch)
    return ''.join(r)
```

---

## Winner

**GPT-5.4** (openrouter)
- Winner Score: 9.14
- Matrix Average: 7.23
- Total Judgments: 63

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | GPT-5.4 | openrouter | 9.14 | 9 |
| 2 | Claude Sonnet 4.6 | openrouter | 8.92 | 8 |
| 3 | GPT-OSS-120B | OpenAI | 8.89 | 8 |
| 4 | Claude Opus 4.6 | openrouter | 8.61 | 9 |
| 5 | Gemini 3 Flash Preview | Google | 8.33 | 9 |
| 6 | DeepSeek V4 | openrouter | 7.13 | 7 |
| 7 | MiMo-V2-Flash | Xiaomi | 4.86 | 5 |
| 8 | Gemini 3.1 Pro | openrouter | 1.96 | 8 |

---

## 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 | — | 8.4 | 0.3 | 8.6 | · | 6.5 | 7.8 | 7.8 | · | · |
| Claude Opus | 10.0 | — | 1.0 | 9.2 | · | 6.9 | 9.2 | 8.9 | · | 1.3 |
| Gemini 3.1 Pro | 8.6 | 8.4 | — | 8.3 | · | 5.3 | · | 9.2 | · | · |
| Claude Sonnet | 9.6 | 8.6 | · | — | · | 7.5 | 9.0 | 8.2 | · | 2.3 |
| Grok 4.20 | 8.8 | 8.8 | 1.9 | 8.8 | — | · | 8.8 | 8.4 | · | 6.0 |
| DeepSeek V4 | 9.0 | 8.4 | 5.0 | 9.2 | · | — | 9.4 | 8.8 | · | 8.1 |
| GPT-OSS-120B | 8.0 | 8.4 | 1.0 | · | · | 6.1 | — | 7.0 | · | · |
| Gemini 3 | 10.0 | 10.0 | 2.2 | 9.8 | · | 9.0 | 9.6 | — | · | · |
| MiniMax M2.5 | 8.8 | 8.4 | 2.5 | 8.6 | · | · | 8.6 | 7.5 | — | 6.8 |
| MiMo-V2-Flash | 9.6 | 8.0 | 1.9 | 9.0 | · | 8.6 | 8.6 | 9.2 | · | — |

---

## 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-007. 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-20260402-123643/results
Full dataset: https://app.themultivac.com/dashboard/export