MiniMax M2.5 vs MiniMax M3 Thinking
MiniMax M2.5 is cheaper on output tokens, while MiniMax M3 Thinking offers a larger context window. Choose MiniMax M2.5 or MiniMax M3 Thinking based on the trade-off between cost, context, and the benchmarks that matter for your use case.
| Spec | MiniMax M2.5 | MiniMax M3 Thinking |
|---|---|---|
| Provider | MiniMax | NanoGPT |
| Input / 1M tokens | $0.27 | $0.23 |
| Output / 1M tokens | $0.95 | $0.96 |
| Context window | 205K | 512K |
| Parameters | — | — |
| Open weights | Yes | Yes |
| Released | Feb 2026 | Jun 2026 |
| LiveBench score | 60.3 | 70.0 |
| SWE-bench Verified score | 75.8 | — |
| Epoch Capabilities Index score | 147 | — |
| LMArena (Chatbot Arena) Elo score | 1359 | — |
| LMArena WebDev Elo score | 1386 | — |
FAQ
Is MiniMax M2.5 or MiniMax M3 Thinking cheaper?
MiniMax M2.5 is cheaper on output tokens ($0.95 vs $0.96 per 1M).
Which has the larger context window, MiniMax M2.5 or MiniMax M3 Thinking?
MiniMax M3 Thinking has the larger context window (512K tokens).
Pricing is indicative — confirm with the provider before production use. Highlighted values indicate the better figure for that row.