MiniMax-M2.7 (229B MoE) Architecture Parameters, Benchmarks & VRAM
229B total / ~10B active high-throughput coding & agentic MoE.
🧬 MiniMax-M2.7 (229B MoE) Architecture & Tensor Specifications
Verified directly against HuggingFace (MiniMaxAI/MiniMax-M2.7) config.json & Model Card
1. Backbone & Dimensions
minimax_m2- Total / Active Params
- 229B / 10B
- Hidden Size (d_model)
- 3,072
- FFN Intermediate Size
- 1,536
- MoE Experts
- 256 Routed · Top-8
- Activation / Norm
- SiLU (SwiGLU) · ε=0.000001
- Native Tensor Dtype
- BF16 / FP8
2. Attention & Layer Layout
GQA- Num Hidden Layers
- 62 Layers
- Hidden Layer Layout
- 62-Layer Sparse MoE (256 Experts, Top-8 Active)
- Attention Heads (Q / KV)
- 48 Q / 8 KV (6:1 GQA)
- Head Dimension
- 128
3. Context, RoPE & Tokenizer
Tokenizer- Native Context Length
- 204,800 tokens
- RoPE Configuration
- RoPE (原生 204,800 Context)
- Vocab Size
- 200,064 (Untied)
📋 MiniMax-M2.7 (229B MoE) Official API & System Card Specifications
Verified against official MiniMax developer documentation & pricing
1. Identity & Capabilities
API Spec- API Model ID
- MiniMax-M2.7
- Release Date
- 2026-03-18
- Weight License
- Open-Weight (API Spec)
- Native Reasoning
- ✅ Supported
- Function Calling & JSON Schema
- ✅ 原生支持
- Batch API 异步批处理
- ✅ Supported (50% off)
2. Context, Output & Modalities
Limits- Max Context Window
- 200,000 tokens (200K)
- Max Output Tokens
- 131,000 tokens
- Input Modalities
- Text
- Output Modalities
- Text
- Direct China Access
- ✅ Direct CN + RMB
- Deployment Mode
- Open-Weight + Cloud API
3. Official Pricing & Cache
CNY / 1M- Input / 1M Tokens
- ¥ 2.1
- Cached Input / 1M
- ¥ 0.42 -80%
- Output / 1M Tokens
- ¥ 8.5
- Tiered Pricing Notes
- MiniMax: Official MiniMax CN API endpoint
🎛️ Official Sampling & Reasoning Parameters
Recommended generation settings & API parameter constraints
📊 MiniMax-M2.7 (229B MoE) Full Benchmark Scorecard
8 public benchmarks · ranks among 154 models · synced 2026-11-20
💻 Coding 5 items
A multilingual benchmark for issue resolving in software engineering that covers Java, TypeScript, JavaScript, Go, Rust, C, and C++.
Terminal-Bench 2.0 is an updated benchmark for testing AI agents' tool use ability to operate a computer via terminal.
SWE-Bench Pro is an advanced version of SWE-Bench that evaluates language models on complex, real-world software engineering tasks requiring extended reasoning and multi-step problem solving.
A multilingual benchmark for issue resolving that evaluates Large Language Models' ability to resolve software issues across diverse programming ecosystems.
NL2Repo evaluates long-horizon coding capabilities including repository-level understanding, where models must generate or modify code across entire repositories from natural language specifications.
🤖 Agents & Tools 2 items
Tool Decathlon is a comprehensive benchmark for evaluating AI agents' ability to use multiple tools across diverse task categories. It measures proficiency in tool selection, sequencing, and execution across ten different tool-use scenarios.
Finance Agent v2 is an agentic financial-analysis benchmark from Vals that evaluates models on real-world finance workflows, measuring their ability to retrieve and reason over financial documents, perform multi-step calculations, and produce accurate analyses.
Source: ZeroEval / llm-stats.com (✓ = third-party verified, others vendor-reported).
⚡ MiniMax-M2.7 (229B MoE) Interactive VRAM & Token Speed Calculator
Physical 512 GB · Load 31.7%
MoE: reads only 6.06 GB active weights/step
Enable Q8/Q4 KV quantization to scale batch concurrency
OLLAMA_CONTEXT_LENGTH=16384 ollama run minimax-m2-7-229bMac Studio M5 Ultra 512G Fits MiniMax-M2.7 (229B MoE) · Cost & TCO
Approx. $9,499 (~$19/GB VRAM) · 190W TDP
📊 MiniMax-M2.7 (229B MoE) Quantization VRAM Breakdown (16K Context)
| Quantization | Bits/Weight | Weights | KV Cache (16K) | Total VRAM | Quality Note |
|---|---|---|---|---|---|
| FP16 / BF16 | 16 bpw | 458 GB | 4.16 GB | 471.77 GB | Lossless Full Precision |
| FP8 / Q8_0 | 8.5 bpw | 243.31 GB | 4.16 GB | 252.79 GB | 99.9% Near-Lossless |
| GGUF Q6_K | 6.56 bpw | 187.78 GB | 4.16 GB | 196.15 GB | Ultra High Quality |
| GGUF Q5_K_M | 5.69 bpw | 162.88 GB | 4.16 GB | 170.74 GB | High Balanced Quality |
| Q4_K_M / AWQ | 4.85 bpw | 138.83 GB | 4.16 GB | 146.22 GB | Sweet Spot (Recommended) |
| GGUF IQ4_XS | 4.25 bpw | 121.66 GB | 4.16 GB | 128.7 GB | Compact 4-bit |
| GGUF Q3_K_M | 3.91 bpw | 111.92 GB | 4.16 GB | 118.77 GB | Moderate Perplexity Loss |
| GGUF Q2_K | 3.15 bpw | 90.17 GB | 4.16 GB | 96.58 GB | Extreme Compression |
🖥️ GPU & Apple Silicon Compatibility for MiniMax-M2.7 (229B MoE) (Q4_K_M)
Frequently Asked Questions
What are the exact architecture parameters of MiniMax-M2.7 (229B MoE)?
MiniMax-M2.7 (229B MoE) uses MiniMaxM2ForCausalLM with 62 layers (62-Layer Sparse MoE (256 Experts, Top-8 Active)), hidden_size=3072, 48 Q heads / 8 KV heads (head_dim=128), and vocab_size=200,064.
How much VRAM does MiniMax-M2.7 (229B MoE) require?
At 4-bit quantization (Q4_K_M, 16K context), MiniMax-M2.7 (229B MoE) requires 146.22 GB VRAM. Full FP16 requires 471.77 GB.