MiniMax M2.5 229B
MiniMax M2.5 229B needs roughly 130.6GB VRAM at Q4_K_M quantization (515.3GB at FP16). 14 GPUs we track can run it fully in VRAM at 8k context.
MiniMax229B params10B active (MoE)192k contextMiniMaxCommercial use ok
VRAM at each quantization
Assumes 8k context. KV cache grows linearly with context length.
| Quant | Weights | KV cache | Total |
|---|---|---|---|
| FP16 | 458.0 GB | 2.08 GB | 515.3 GB |
| Q8 | 229.0 GB | 2.08 GB | 258.8 GB |
| Q6_K | 171.8 GB | 2.08 GB | 194.7 GB |
| Q5_K_M | 143.1 GB | 2.08 GB | 162.6 GB |
| Q4_K_M | 114.5 GB | 2.08 GB | 130.6 GB |
| Q3_K_M | 91.6 GB | 2.08 GB | 104.9 GB |
| Q2_K | 68.7 GB | 2.08 GB | 79.3 GB |
Benchmarks
GPUs that run MiniMax M2.5 229B natively (14)
- NVIDIA DGX Spark (128GB)Q3_K_M · 75.1 t/s
- AMD Instinct MI300XQ5_K_M · 932.8 t/s
- AMD Strix Halo (128GB)Q3_K_M · 70.4 t/s
- AMD Strix Halo (96GB)Q2_K · 93.9 t/s
- Apple M4 Ultra (384GB)Q8 · 120.1 t/s
- Apple M4 Ultra (192GB)Q5_K_M · 192.2 t/s
- Apple M4 Max (128GB)Q3_K_M · 150.2 t/s
- Apple M4 Max (96GB)Q2_K · 200.2 t/s
- Apple M3 Max (128GB)Q3_K_M · 110 t/s
- Apple M3 Max (96GB)Q2_K · 146.7 t/s
- Apple M2 Ultra (384GB)Q8 · 88 t/s
- Apple M2 Ultra (192GB)Q5_K_M · 140.8 t/s
- Apple M2 Max (96GB)Q2_K · 146.7 t/s
- Apple M1 Ultra (128GB)Q3_K_M · 220 t/s
Plus 2 GPUs that run it with CPU offload (slower)
- NVIDIA H100 80GBQ2_K · 279.2 t/s
- NVIDIA A100 80GBQ2_K · 169.9 t/s
Notes
Coding and agentic focus — SWE-Bench Verified 80.2%. 256 experts, 8 active per token.