# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260315-054647
**Date:** Mar 15, 2026
**Category:** code
**Question ID:** EVAL-20260315-054647

---

## Question

Implement an LRU cache with per-key TTL (time-to-live) support. Requirements: O(1) get/put, thread-safe, lazy expiration (don't use background threads), configurable max size, eviction callback, and cache hit/miss statistics. Include comprehensive tests.

---

## Winner

**Qwen 3 8B** (openrouter)
- Winner Score: 9.23
- Matrix Average: 8.17
- Total Judgments: 66

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | Qwen 3 8B | openrouter | 9.23 | 8 |
| 2 | Gemma 3 27B | openrouter | 9.12 | 8 |
| 3 | Llama 4 Scout | openrouter | 8.57 | 7 |
| 4 | Granite 4.0 Micro | openrouter | 8.40 | 8 |
| 5 | Devstral Small | openrouter | 8.29 | 8 |
| 6 | Qwen 3 32B | openrouter | 8.15 | 1 |
| 7 | Kimi K2.5 | openrouter | 8.15 | 1 |
| 8 | Phi-4 14B | openrouter | 8.14 | 9 |
| 9 | Llama 3.1 8B | openrouter | 7.45 | 8 |
| 10 | Mistral Nemo 12B | openrouter | 6.18 | 8 |

---

## 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 | — | · | 8.1 | 8.3 | 8.4 | 8.3 | 7.0 | 8.4 | 3.8 | 4.8 |
| Kimi K2.5 | · | — | · | · | · | 5.4 | · | · | · | · |
| Devstral Small | · | · | — | 9.6 | 8.2 | 8.6 | 8.6 | 10.0 | 5.8 | 7.4 |
| Gemma 3 27B | · | · | 8.3 | — | 8.6 | 8.4 | 8.8 | 9.4 | 6.6 | 8.4 |
| Llama 4 Scout | · | · | 8.4 | 9.6 | — | 8.8 | 8.8 | 9.6 | 5.6 | 8.6 |
| Phi-4 14B | · | · | 8.3 | 9.6 | 8.4 | — | 9.0 | 9.8 | 6.6 | 8.1 |
| Granite 4.0 | 8.2 | 8.2 | 8.8 | 8.8 | · | 8.8 | — | 8.8 | 8.8 | 8.8 |
| Qwen 3 8B | · | · | 8.3 | 9.2 | 8.5 | 7.2 | 6.8 | — | 3.6 | 5.2 |
| Mistral Nemo | · | · | 7.7 | 8.7 | 8.3 | 8.3 | 8.6 | 8.3 | — | 8.4 |
| Llama 3.1 8B | · | · | 8.6 | 9.3 | 9.6 | 9.6 | 9.6 | 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-20260315-054647. 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-20260315-054647/results
Full dataset: https://app.themultivac.com/dashboard/export