Model reference · open weights
minimax-m2-tiny is an open-weight language model from PrimeIntellect. minimax-m2-tiny (FP32) weighs 504 MB; the smallest configuration that runs it is RTX 3060 12 GB.
Summary of the PrimeIntellect/minimax-m2-tiny model card, 2026-10-01
What it is
| Released by | PrimeIntellect |
|---|---|
| Released | 2026-02-18 |
| Parameters | 252M |
| VRAM | 504 MB for the weights |
What it runs on
| Card | Requests at once | Context max | Memory |
|---|---|---|---|
| 4K each | |||
| RTX 3060 12 GB | 207 | all 4K | 11.6 GB |
| RTX 4060 Ti 16 GB | 283 | all 4K | 15.4 GB |
| RTX 3090 24 GB | 442 | all 4K | 23.4 GB |
| RTX 4090 24 GB | 441 | all 4K | 23.4 GB |
| RTX 5090 32 GB | 593 | all 4K | 31.0 GB |
| L40S 48 GB | 850 | all 4K | 44.0 GB |
| A100 80 GB | 1000+ | all 4K | 78.2 GB |
| H100 80 GB | 1000+ | all 4K | 78.1 GB |
| RTX PRO 6000 Blackwell 96 GB | 1000+ | all 4K | 93.8 GB |
| DGX Spark (GB10) 128 GB unified | 1000+ | all 4K | 107 GB |
| H200 141 GB | 1000+ | all 4K | 138 GB |
| B200 180 GB | 1000+ | all 4K | 176 GB |
| Requests at once | 4K tokens each |
|---|---|
| 1 | 1.2 GB |
| 5 | 1.4 GB |
| 8 | 1.6 GB |
| 16 | 2.0 GB |
| 32 | 2.8 GB |
| 64 | 4.4 GB |
One card, with vLLM's small-card settings.
From the model card
A small (~252M parameter) MiniMax M2 MoE model for testing only. It is generally compatible with vLLM and HuggingFace Transformers but is meant to be used with prime-rl.
This model has random weights (no SFT warmup yet due to a chat template tokenization issue with MiniMax's tokenizer).
uv run rl @ configs/ci/integration/rl_moe/minimax_m2.toml
See the Testing MoE at Small Scale guide for full instructions.
| Parameter | Value |
|---|---|
| Hidden size | 512 |
| Layers | 12 |
| Experts | 8 |
| Active experts | 4 |
| Parameters | ~252M |
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.