Hy3vsMiniMax M3
Across 4 shared benchmarks, Hy3 leads overall: Hy3 wins 3, MiniMax M3 wins 1, with 0 ties and an average score difference of +2.55.
Hy3
腾讯AI实验室 · 2026-07-06 · Reasoning model
MiniMax M3
MiniMaxAI · 2026-06-01 · Multimodal model
Hy33 wins(75%)(25%)1 winMiniMax M3
Benchmark scores
Grouped by capability, sorted by largest gap within each. 4 shared benchmarks.
AI Agent - Tool Usage
Hy3 2/2| Benchmark | Hy3 | MiniMax M3 | Diff |
|---|---|---|---|
| Terminal-Bench 2.1 | 71.7030 / 44Thinking High (With Tools) | 6635 / 44Thinking (With Tools) | +5.70 |
| MCP-Atlas | 79.109 / 38Thinking High (With Tools) | 74.2022 / 38Thinking (With Tools) | +4.90 |
AI Agent - Information Search
Hy3 1/1| Benchmark | Hy3 | MiniMax M3 | Diff |
|---|---|---|---|
| BrowseComp | 84.2010 / 54Thinking High (With Tools + Internet) | 83.5012 / 54Thinking (With Tools + Internet) | +0.70 |
Coding and Software Engineer
MiniMax M3 1/1| Benchmark | Hy3 | MiniMax M3 | Diff |
|---|---|---|---|
| SWE-Bench Pro - Public | 57.9018 / 57Thinking High (With Tools) | 5913 / 57Thinking (With Tools) | -1.10 |
Specs
| Field | Hy3 | MiniMax M3 |
|---|---|---|
| Publisher | 腾讯AI实验室 | MiniMaxAI |
| Release date | 2026-07-06 | 2026-06-01 |
| Model type | Reasoning model | Multimodal model |
| Architecture | MoE | MoE |
| Parameters | 295B | 428B |
| Context length | 256K | 1M |
| Max output | Not available | 512K |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | Hy3 | MiniMax M3 |
|---|---|---|
| Text input | ¥1.2 / 1M tokens | ¥2.1 / 1M tokens |
| Text output | ¥4 / 1M tokens | ¥8.4 / 1M tokens |
| Cache read | ¥0.4 / 1M tokens | ¥0.42 / 1M tokens |
Summary
- Hy3leads in:AI Agent - Tool Usage (2/2), AI Agent - Information Search (1/1)
- MiniMax M3leads in:Coding and Software Engineer (1/1)
On average across the 4 shared benchmarks, Hy3 scores 2.55 higher.
Largest single-benchmark gap: Terminal-Bench 2.1 — Hy3 71.70 vs MiniMax M3 66 (+5.70).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.