MiniMax M3vsStep 3.7 Flash
Across 3 shared benchmarks, MiniMax M3 leads overall: MiniMax M3 wins 3, Step 3.7 Flash wins 0, with 0 ties and an average score difference of +5.63.
MiniMax M3
MiniMaxAI · 2026-06-01 · Multimodal model
Step 3.7 Flash
StepFunAI · 2026-05-29 · Reasoning model
MiniMax M33 wins(100%)(0%)0 winsStep 3.7 Flash
Benchmark scores
Grouped by capability, sorted by largest gap within each. 3 shared benchmarks.
AI Agent - Information Search
MiniMax M3 1/1| Benchmark | MiniMax M3 | Step 3.7 Flash | Diff |
|---|---|---|---|
| BrowseComp | 83.5012 / 54Thinking (With Tools + Internet) | 75.8225 / 54Thinking (With Tools) | +7.68 |
AI Agent - Tool Usage
MiniMax M3 1/1| Benchmark | MiniMax M3 | Step 3.7 Flash | Diff |
|---|---|---|---|
| Terminal-Bench 2.1 | 6635 / 44Thinking (With Tools) | 59.5037 / 44Thinking (With Tools) | +6.50 |
Coding and Software Engineer
MiniMax M3 1/1| Benchmark | MiniMax M3 | Step 3.7 Flash | Diff |
|---|---|---|---|
| SWE-Bench Pro - Public | 5913 / 57Thinking (With Tools) | 56.3025 / 57Thinking (With Tools) | +2.70 |
Specs
| Field | MiniMax M3 | Step 3.7 Flash |
|---|---|---|
| Publisher | MiniMaxAI | StepFunAI |
| Release date | 2026-06-01 | 2026-05-29 |
| Model type | Multimodal model | Reasoning model |
| Architecture | MoE | MoE |
| Parameters | 428B | 198B |
| Context length | 1M | 256K |
| Max output | 512K | Not available |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | MiniMax M3 | Step 3.7 Flash |
|---|---|---|
| Text input | ¥2.1 / 1M tokens | ¥1.35 / 1M tokens |
| Text output | ¥8.4 / 1M tokens | ¥8.1 / 1M tokens |
| Cache read | ¥0.42 / 1M tokens | ¥0.27 / 1M tokens |
Summary
- MiniMax M3leads in:AI Agent - Information Search (1/1), AI Agent - Tool Usage (1/1), Coding and Software Engineer (1/1)
On average across the 3 shared benchmarks, MiniMax M3 scores 5.63 higher.
Largest single-benchmark gap: BrowseComp — MiniMax M3 83.50 vs Step 3.7 Flash 75.82 (+7.68).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.