MiniMax M3vsQwen3.7 Max
Across 4 shared benchmarks, Qwen3.7 Max leads overall: MiniMax M3 wins 0, Qwen3.7 Max wins 4, with 0 ties and an average score difference of -5.27.
MiniMax M3
MiniMaxAI · 2026-06-01 · Multimodal model
Qwen3.7 Max
阿里巴巴 · 2026-05-20 · Reasoning model
MiniMax M30 wins(0%)(100%)4 winsQwen3.7 Max
Benchmark scores
Grouped by capability, sorted by largest gap within each. 4 shared benchmarks.
Coding and Software Engineer
Qwen3.7 Max 2/2| Benchmark | MiniMax M3 | Qwen3.7 Max | Diff |
|---|---|---|---|
| Text Arena (Coding) | 1,52814 / 35Normal (No Tools) | 1,54111 / 35Normal (No Tools) | -13.02 |
| SWE-Bench Pro - Public | 5913 / 57Thinking (With Tools) | 60.6012 / 57Thinking (With Tools) | -1.60 |
AI Agent - Tool Usage
Qwen3.7 Max 1/1| Benchmark | MiniMax M3 | Qwen3.7 Max | Diff |
|---|---|---|---|
| MCP-Atlas | 74.2022 / 38Thinking (With Tools) | 76.4016 / 38Thinking (With Tools) | -2.20 |
General Knowledge
Qwen3.7 Max 1/1| Benchmark | MiniMax M3 | Qwen3.7 Max | Diff |
|---|---|---|---|
| LiveBench | 70.0240 / 115Deep Thinking (No Tools) | 74.2921 / 115Deep Thinking (No Tools) | -4.27 |
Specs
| Field | MiniMax M3 | Qwen3.7 Max |
|---|---|---|
| Publisher | MiniMaxAI | 阿里巴巴 |
| Release date | 2026-06-01 | 2026-05-20 |
| Model type | Multimodal model | Reasoning model |
| Architecture | MoE | Dense |
| Parameters | 428B | Not available |
| Context length | 1M | 1M |
| Max output | 512K | 64K |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | MiniMax M3 | Qwen3.7 Max |
|---|---|---|
| Text input | ¥2.1 / 1M tokens | ¥12 / 1M tokens |
| Text output | ¥8.4 / 1M tokens | ¥36 / 1M tokens |
| Cache read | ¥0.42 / 1M tokens | Not public |
Summary
- Qwen3.7 Maxleads in:Coding and Software Engineer (2/2), AI Agent - Tool Usage (1/1), General Knowledge (1/1)
On average across the 4 shared benchmarks, Qwen3.7 Max scores 5.27 higher.
Largest single-benchmark gap: Text Arena (Coding) — MiniMax M3 1,528 vs Qwen3.7 Max 1,541 (-13.02).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.