# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260315-034158
**Date:** Mar 15, 2026
**Category:** code
**Question ID:** EVAL-20260315-034158

---

## Question

This SQL query takes 45 seconds on a table with 10M rows. Rewrite it to run in under 1 second. Explain your optimization strategy.

```sql
SELECT u.name, u.email,
       (SELECT COUNT(*) FROM orders o WHERE o.user_id = u.id) as order_count,
       (SELECT SUM(amount) FROM orders o WHERE o.user_id = u.id) as total_spent,
       (SELECT MAX(created_at) FROM orders o WHERE o.user_id = u.id) as last_order
FROM users u
WHERE u.created_at > '2024-01-01'
AND (SELECT COUNT(*) FROM orders o WHERE o.user_id = u.id) > 5
ORDER BY total_spent DESC
LIMIT 100;
```

Assume standard B-tree indexes on primary keys only. What indexes would you add?

---

## Winner

**Qwen 3 32B** (openrouter)
- Winner Score: 9.66
- Matrix Average: 8.70
- Total Judgments: 76

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | Qwen 3 8B | openrouter | 9.66 | 7 |
| 2 | Qwen 3 32B | openrouter | 9.66 | 8 |
| 3 | Gemma 3 27B | openrouter | 9.57 | 8 |
| 4 | Mistral Nemo 12B | openrouter | 9.17 | 8 |
| 5 | Devstral Small | openrouter | 9.03 | 8 |
| 6 | Granite 4.0 Micro | openrouter | 8.91 | 8 |
| 7 | Llama 4 Scout | openrouter | 8.80 | 8 |
| 8 | Phi-4 14B | openrouter | 8.54 | 9 |
| 9 | Llama 3.1 8B | openrouter | 8.12 | 9 |
| 10 | Kimi K2.5 | openrouter | 5.57 | 3 |

---

## 10×10 Judgment Matrix

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

| Judge ↓ / Resp → | Qwen 3 32B | Kimi K2.5 | Devstral Small | Gemma 3 27B | Llama 4 Scout | Phi-4 14B | Granite 4.0 | Qwen 3 8B | Mistral Nemo | Llama 3.1 8B |
|---|---|---|---|---|---|---|---|---|---|---|
| Qwen 3 32B | — | · | 7.8 | 9.2 | 8.0 | 8.4 | 9.0 | 9.6 | 8.6 | 8.0 |
| Kimi K2.5 | · | — | · | · | · | 6.2 | · | · | · | 4.0 |
| Devstral Small | 10.0 | · | — | 10.0 | 8.6 | 9.6 | 9.8 | 10.0 | 9.6 | 8.6 |
| Gemma 3 27B | 9.8 | 1.0 | 9.6 | — | 8.9 | 9.4 | 8.9 | 9.8 | 9.8 | 8.4 |
| Llama 4 Scout | 9.8 | · | 9.6 | 9.8 | — | 8.8 | 8.8 | 9.8 | 9.6 | 8.4 |
| Phi-4 14B | 9.8 | 8.2 | 9.8 | 10.0 | 8.6 | — | 8.8 | 10.0 | 9.8 | 9.2 |
| Granite 4.0 | 8.8 | 7.5 | 8.8 | 8.8 | 8.4 | 8.8 | — | 8.8 | 8.8 | 8.8 |
| Qwen 3 8B | 9.4 | · | 9.4 | 9.4 | 9.2 | 9.4 | 8.8 | — | 8.6 | 8.7 |
| Mistral Nemo | 10.0 | · | 8.1 | 9.8 | 9.6 | 7.7 | 8.3 | · | — | 9.1 |
| Llama 3.1 8B | 9.6 | · | 9.2 | 9.6 | 9.1 | 8.6 | 8.8 | 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: EVAL-20260315-034158. 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-20260315-034158/results
Full dataset: https://app.themultivac.com/dashboard/export