M2
#141 in Frontier-Spraakmodelleminimax · siet 2025-10-27 · 4× · tolest 30. Juni 2026
8
Momentum
MiniMax M2 is an open-source language model by Chinese AI company MiniMax, designed for agent and coding workflows. It uses a Mixture-of-Experts (MoE) architecture with 230 billion total parameters and 10 billion active parameters per token. The model features interleaved thinking (chain-of-thought) and is optimized for tool use (Shell, Browser, Python, MCP tools) and multi-step agentic pipelines. Released under the MIT license, model weights are freely available on Hugging Face.
Momentum-Verloop
04.04.03.07.
Features
| Context Window (Tokens) | 204,800 tokens (max. output: 196,608 tokens) |
| Multimodal (Image/Video) | No – M2 supports text input and text output only; no image or video input |
| NLP Benchmark Score | Artificial Analysis Intelligence Index: 28 (estimated; above average for open-weight models of similar size, median: 25); inference speed: ~108 tokens/s |
| Price per 1M Input Tokens | $0.30 / 1M input tokens; $1.20 / 1M output tokens (official MiniMax API pricing) |
| Price Tier (Monthly) | MiniMax Agent: Free (Lightning Mode, 1,000 credits for new customers), Basic $19/month (~10,000 credits), Pro $69/month (~40,000 credits) |