What is MiniMax? The Shanghai AI Lab Rivaling Claude at 1/50th the Cost
MiniMax is a Shanghai-based AI company that IPO’d in Hong Kong in January 2026. Their M2.7 model delivers roughly 90% of Claude Opus 4.6’s coding quality at 1/50th the price — $0.30 per million input tokens vs Claude’s $15.
Key facts
- Founded: 2021, Shanghai
- IPO: January 2026 (Hong Kong Stock Exchange)
- Flagship model: MiniMax M2.7 (230B MoE, 10B active)
- Pricing: $0.30/$1.20 per 1M tokens (input/output)
- SWE-Pro score: 56.22% (competitive with frontier models)
- Speed: ~100 tokens/second
The model lineup
| Model | Released | Params | Best for | Price (input) |
|---|---|---|---|---|
| M2.7 | Mar 2026 | 230B MoE (10B active) | Agentic coding, reasoning | $0.30/1M |
| M2.5 | Feb 2026 | 230B MoE | Coding (80.2% SWE-bench) | $0.15/1M |
| M2 | Earlier | — | General tasks | Cheaper |
Why MiniMax matters
For developers: M2.7 at $0.30/1M input tokens is 50x cheaper than Claude Opus and 33x cheaper than GPT-5.4. For routine coding tasks, the quality difference is barely noticeable.
The self-evolving angle: M2.7 is marketed as “the first AI model to actively participate in its own evolutionary process.” It uses multi-agent collaboration to plan, execute, and refine complex tasks.
Architecture: 230B total parameters with only 10B active per token (4.3% activation rate). 256 local experts, 8 activated per token, 62 layers. This extreme sparsity is how they keep costs so low.
How it compares
| MiniMax M2.7 | Claude Opus 4.6 | DeepSeek Chat | GLM-5.1 | |
|---|---|---|---|---|
| Input price | $0.30/1M | $15.00/1M | $0.27/1M | $1.00/1M |
| Quality | ~90% of Opus | Best | ~85% of Opus | ~95% of Opus |
| Speed | 100 tok/s | 50 tok/s | 60 tok/s | 55 tok/s |
| Context | 200K | 200K | 128K | 200K |
MiniMax and DeepSeek are the two cheapest frontier-class options. DeepSeek is slightly cheaper, MiniMax is slightly faster.
How to access MiniMax
- MiniMax API — Direct access at api.minimax.chat
- OpenRouter —
minimax/minimax-m2.7(one API key for all models) - Aider —
aider --model openrouter/minimax/minimax-m2.7 - OpenCode — via OpenRouter provider
See our MiniMax API guide for code examples and our MiniMax M2.7 Complete Guide for full details.
How to use MiniMax
Available through:
- MiniMax API directly
- OpenRouter —
minimax/minimax-m2.7 - Aider — via OpenRouter
- OpenCode — via API
See our MiniMax M2.7 Complete Guide for setup instructions.
FAQ
Is MiniMax M2.7 good enough to replace Claude for coding?
For routine coding tasks (generating functions, writing boilerplate, fixing bugs), M2.7 delivers roughly 90% of Claude Opus’s quality at 1/50th the price. For complex architectural decisions, multi-step refactoring, or tasks requiring deep reasoning, Claude still has a meaningful edge. Many developers use MiniMax for high-volume tasks and Claude for critical ones.
How do I access MiniMax models?
The easiest way is through OpenRouter using the model ID minimax/minimax-m2.7. This works with any OpenAI-compatible tool including Aider and OpenCode. You can also use MiniMax’s direct API at api.minimax.chat.
What does “self-evolving” mean for M2.7?
MiniMax markets M2.7 as “the first AI model to actively participate in its own evolutionary process.” In practice, this means the model uses multi-agent collaboration during training and inference — multiple specialized agents plan, execute, and refine complex tasks together, iteratively improving outputs beyond what a single forward pass produces.
Related: MiniMax M2.7 Complete Guide · MiniMax M2.7 vs Claude vs DeepSeek · Best Open-Source Coding Models 2026