# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260403-151843
**Date:** Apr 03, 2026
**Category:** code
**Question ID:** SLM-001

---

## Question

Summarize this 500-word passage in exactly 50 words while retaining all key claims: [Passage about climate change policy]. This tests whether small models can do precise length-constrained summarization.

---

## Winner

**Qwen 3 32B** (openrouter)
- Winner Score: 7.87
- Matrix Average: 5.81
- Total Judgments: 56

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | Qwen 3 8B | openrouter | 8.38 | 6 |
| 2 | Qwen 3 32B | openrouter | 7.87 | 7 |
| 3 | Devstral Small | openrouter | 7.49 | 7 |
| 4 | Llama 4 Scout | openrouter | 7.31 | 7 |
| 5 | Nemotron 3 Super | openrouter | 5.41 | 7 |
| 6 | Kimi K2.5 | openrouter | 4.83 | 8 |
| 7 | Granite 4.0 Micro | openrouter | 4.29 | 7 |
| 8 | Gemma 3n 4B | openrouter | 3.68 | 3 |
| 9 | Gemma 3 27B | openrouter | 3.06 | 4 |

---

## 10×10 Judgment Matrix

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

| Judge ↓ / Resp → | Qwen 3 32B | Nemotron 3 | Devstral Small | Gemma 3 27B | Llama 4 Scout | Granite 4.0 | Gemma 3n 4B | Qwen 3 8B | Kimi K2.5 |
|---|---|---|---|---|---|---|---|---|---|
| Qwen 3 32B | — | 4.8 | 8.1 | 2.0 | 8.3 | 4.5 | · | 8.3 | 7.0 |
| Nemotron 3 | 5.2 | — | 5.3 | 1.8 | 3.2 | 1.8 | 0.4 | 7.8 | 5.2 |
| Devstral Small | 9.3 | 2.0 | — | · | 7.8 | 7.5 | · | 9.3 | 2.0 |
| Gemma 3 27B | 8.3 | 6.0 | 8.1 | — | 8.1 | 4.8 | 2.4 | 8.4 | 7.0 |
| Llama 4 Scout | 8.3 | 6.0 | 7.7 | 1.0 | — | 5.3 | · | 8.3 | 1.0 |
| Granite 4.0 | 8.1 | 8.1 | 8.3 | 7.5 | 8.1 | — | 8.3 | 8.3 | 5.0 |
| Gemma 3n 4B | 8.1 | 4.5 | 7.3 | · | 8.1 | 3.4 | — | · | 4.5 |
| Qwen 3 8B | 8.1 | 6.5 | 7.8 | · | 7.8 | 2.9 | · | — | 7.0 |
| Kimi K2.5 | · | · | · | · | · | · | · | · | — |

---

## 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: SLM-001. 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-20260403-151843/results
Full dataset: https://app.themultivac.com/dashboard/export