Core Conclusion
MiniMax M3 is entering release countdown. Community messages show this model is expected to release in May, and MiniMax has already demonstrated two key trends in the previous version M2.7: aggressive pricing ($0.3 per million input tokens) and Agent capabilities (self-evolving Agent building).
M3’s arrival may add fuel to the already white-hot domestic model competition.
MiniMax’s Evolution Path
From M2.5 to M2.7
| Version | Release Time | Input Price ($/M token) | Core Selling Point |
|---|---|---|---|
| MiniMax 2.5 | Late 2025 | Unreleased (free public beta) | Distillation technology, multimodal |
| MiniMax M2.7 | March 2026 | 0.3 (input) / 2.1 (RMB input price) | SWE-Pro approaching Opus, Agent self-evolution |
| MiniMax M3 | Expected May 2026 | TBD | TBD |
A key signal from M2.7: the SWE-Pro version approaches Claude Opus in coding Agent capability, but at a fraction of Opus’s price.
Pricing Strategy Analysis
In the March 2026 mainstream model API pricing:
| Model | Input Price ($/M token) |
|---|---|
| Grok | 0.20 |
| MiniMax M2.7 | 0.30 |
| DeepSeek V4 Flash | 0.35 |
| Qwen3.6 Plus | 0.40 |
| Kimi K2.6 | 0.50 |
| Claude Sonnet 4.5 | 3.00 |
| GPT-5.5 | 2.50 |
MiniMax’s pricing strategy is very clear: use extreme pricing to gain market share, then build differentiation through Agent capabilities (rather than pure benchmark scores).
What M3 Might Bring
Performance Upgrade Expectations
Based on MiniMax’s technical route and M2.7’s foundation, M3 may upgrade in the following areas:
- Enhanced Agent capabilities: M2.7 already supports “self-evolving Agent building”, M3 may further strengthen multi-step planning and tool calling
- Context window expansion: from M2.7’s 128K may expand to 256K or higher
- MoE architecture optimization: reduce activation parameter ratio, improve inference speed
- Multimodal integration: may add visual understanding capability
Pricing Prediction
Considering MiniMax’s consistently aggressive pricing strategy, M3’s pricing may follow two paths:
| Path | Input Price Prediction | Logic |
|---|---|---|
| Maintain aggressive | $0.25-$0.30 | Continue gaining share through pricing |
| Slight increase | $0.40-$0.50 | If performance significantly improves, may moderately adjust |
The more likely scenario is maintaining low prices, because MiniMax’s core competitiveness is cost-effectiveness.
Competitive Landscape Assessment
MiniMax’s Position Among Domestic Models
| Dimension | MiniMax | Qwen | Kimi | DeepSeek | GLM |
|---|---|---|---|---|---|
| Price competitiveness | ⭐⭐⭐⭐⭐ | ⭐⭐⭐ | ⭐⭐⭐ | ⭐⭐⭐⭐ | ⭐⭐ |
| Coding ability | ⭐⭐⭐⭐ | ⭐⭐⭐⭐⭐ | ⭐⭐⭐ | ⭐⭐⭐⭐ | ⭐⭐⭐⭐⭐ |
| Agent capability | ⭐⭐⭐⭐ | ⭐⭐⭐ | ⭐⭐⭐ | ⭐⭐⭐⭐ | ⭐⭐⭐ |
| Chinese ability | ⭐⭐⭐ | ⭐⭐⭐⭐⭐ | ⭐⭐⭐⭐ | ⭐⭐⭐ | ⭐⭐⭐⭐ |
| Ecosystem building | ⭐⭐ | ⭐⭐⭐⭐⭐ | ⭐⭐⭐ | ⭐⭐⭐ | ⭐⭐⭐ |
MiniMax’s differentiation lies in: lowest price + outstanding Agent capability. It doesn’t follow Qwen’s all-around route, nor GLM’s coding-specialization route, but chooses a unique “extreme cost-effectiveness + Agent-first” narrow track.
Action Recommendations
For developers:
- If currently using MiniMax M2.7, pay attention to M3’s API compatibility
- Evaluate whether to migrate from other models after M3 release
- Utilize M2.7’s current low-price window to accelerate prototype development
For corporate decision-makers:
- Don’t rush to sign long-term model service contracts before M3 release
- Evaluate whether MiniMax’s Agent capabilities meet business needs
- Pay attention to the impact of price wars on overall API costs after M3 release
For investors:
- MiniMax’s pricing strategy reflects its commercialization pace
- Watch user growth and revenue changes after M3 release
- Pay attention to MiniMax’s investment in Agent ecosystem