MiniMax M1
MiniMax M1 from MiniMax is a hybrid-attention reasoning model for long context, tool use, software engineering, and efficient reinforcement learning.
Model Overview
Capabilities, design details, and architectural traits
MiniMax M1 - hybrid-attention reasoning model
MiniMax M1 is built around hybrid attention and long-context reasoning. Its documented identity is a reasoning model with a large context window, long reasoning output, and agent and software engineering use cases.
| Distinctive trait | What is explicitly documented |
|---|---|
| Hybrid-attention design | The model uses a proprietary hybrid-attention setup with Lightning Attention. |
| Long-context reasoning | It supports a 1 million token context window and 80k token reasoning output. |
| CISPO reinforcement learning | The announcement names CISPO as the RL algorithm used to improve training efficiency. |
| Tool and software focus | The model is described as strong in software engineering, long context, and tool use. |
Attention pattern
The key idea is not generic chat. The model is presented as a reasoning system that combines long input handling with efficient inference on deep context.
Training method
The release also names CISPO as a faster reinforcement learning method. The model is described as using this with Lightning Attention to improve efficiency in training and inference.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | MiniMax | Anthropic | Anthropic |
| Release Date | June 17, 2025 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | Jun 2024 | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 1M | 1M | 1M |
| Pricing (per 1M tokens) | |||
| Input Pricing | $0.55 Best Input Pricing | $4 | $2 |
| Output Pricing | $2.20 Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | text | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 11.7 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | - | - | - |
| Agentic Index | - | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from MiniMax
Other models by MiniMax
Top AI Models
Leading alternatives by intelligence score
