# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260402-193645
**Date:** Apr 02, 2026
**Category:** analysis
**Question ID:** ANALYSIS-014

---

## Question

You're launching an AI API. Competitors charge $0.01-0.03/1K tokens. Your model is 20% better on benchmarks but 40% more expensive to run. (1) Should you price above, at, or below competitors? Analyze each strategy. (2) Design a pricing structure (free tier, usage-based, enterprise). (3) A customer processes 10M tokens/month. Calculate their cost under your pricing vs competitors. (4) At what volume does a dedicated instance become cheaper than per-token pricing?

---

## Winner

**MiniMax M2.5** (openrouter)
- Winner Score: 8.72
- Matrix Average: 7.88
- Total Judgments: 89

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | MiniMax M2.5 | openrouter | 8.72 | 9 |
| 2 | MiMo-V2-Flash | Xiaomi | 8.51 | 9 |
| 3 | Grok 4.20 | openrouter | 8.42 | 9 |
| 4 | GPT-5.4 | openrouter | 8.22 | 9 |
| 5 | Gemini 3 Flash Preview | Google | 8.03 | 9 |
| 6 | Claude Sonnet 4.6 | openrouter | 7.98 | 9 |
| 7 | Claude Opus 4.6 | openrouter | 7.71 | 9 |
| 8 | DeepSeek V4 | openrouter | 7.57 | 9 |
| 9 | GPT-OSS-120B | OpenAI | 7.50 | 8 |
| 10 | Gemini 3.1 Pro | openrouter | 6.12 | 9 |

---

## 10×10 Judgment Matrix

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

| Judge ↓ / Resp → | Gemini 3.1 Pro | Claude Opus | GPT-5.4 | DeepSeek V4 | MiMo-V2-Flash | Claude Sonnet | Grok 4.20 | Gemini 3 | GPT-OSS-120B | MiniMax M2.5 |
|---|---|---|---|---|---|---|---|---|---|---|
| Gemini 3.1 Pro | — | 6.4 | 7.9 | 7.0 | 8.8 | 7.5 | 6.5 | 7.2 | · | 8.9 |
| Claude Opus | 6.0 | — | 7.5 | 6.3 | 7.8 | 8.2 | 8.8 | 7.8 | 7.0 | 7.8 |
| GPT-5.4 | 3.3 | 6.0 | — | 6.7 | 7.2 | 5.8 | 8.4 | 8.0 | 5.3 | 8.4 |
| DeepSeek V4 | 8.6 | 9.4 | 8.8 | — | 9.0 | 9.2 | 9.3 | 9.0 | 9.0 | 9.0 |
| MiMo-V2-Flash | 7.3 | 7.8 | 8.0 | 8.2 | — | 9.3 | 8.6 | 8.6 | 7.8 | 9.2 |
| Claude Sonnet | 6.3 | 8.8 | 8.2 | 7.3 | 8.6 | — | 8.6 | 7.8 | 7.3 | 8.6 |
| Grok 4.20 | 6.5 | 8.7 | 8.8 | 7.0 | 8.4 | 8.4 | — | 7.8 | 8.6 | 8.8 |
| Gemini 3 | 7.3 | 9.6 | 9.6 | 9.6 | 9.6 | 9.6 | 9.8 | — | 8.8 | 9.6 |
| GPT-OSS-120B | 4.3 | 5.9 | 7.7 | 7.6 | 8.6 | 6.5 | 7.8 | 7.8 | — | 8.3 |
| MiniMax M2.5 | 5.8 | 6.8 | 7.6 | 8.6 | 8.8 | 7.4 | 8.2 | 8.6 | 6.1 | — |

---

## 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: ANALYSIS-014. 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-20260402-193645/results
Full dataset: https://app.themultivac.com/dashboard/export