Microsoft models: prices and the hardware each needs
19 models, 1 sold per token by 2 vendors. Vendors are in alphabetical order; the price is the cheapest list price among them, per million tokens.
| Model | Size | Sold per token by | From, input | Output | GPU memory |
|---|---|---|---|---|---|
| maira-2 | 6.88 B | not sold per token | 34.4 GB | ||
| phi-1 | 1.42 B | not sold per token | 3.5 GB | ||
| phi-1_5 | 1.42 B | not sold per token | 3.5 GB | ||
| phi-2 | 2.78 B | not sold per token | 6.9 GB | ||
| Phi-3-medium-4k-instruct | 13.96 B | not sold per token | 34.9 GB | ||
| Phi-3-mini-128k-instruct | 3.82 B | not sold per token | 9.5 GB | ||
| Phi-3-mini-4k-instruct | 3.82 B | not sold per token | 9.5 GB | ||
| Phi-3-small-8k-instruct | 7.39 B | not sold per token | 18.5 GB | ||
| Phi-3-vision-128k-instruct | 4.15 B | not sold per token | 10.4 GB | ||
| Phi-3.5-mini-instruct | 3.82 B | not sold per token | 9.5 GB | ||
| Phi-3.5-MoE-instruct | 41.87 B | not sold per token | 104.7 GB | ||
| Phi-3.5-vision-instruct | 4.15 B | not sold per token | 10.4 GB | ||
| Phi-4 | 14 B | Azure, DeepInfra | $0.07 | $0.14 | 36.6 GB |
| Phi-4-mini-instruct | 3.84 B | not sold per token | 9.6 GB | ||
| Phi-4-mini-reasoning | 3.84 B | not sold per token | 9.6 GB | ||
| Phi-4-multimodal-instruct | 5.57 B | not sold per token | 13.9 GB | ||
| Phi-4-reasoning | 14.66 B | not sold per token | 36.6 GB | ||
| Phi-mini-MoE-instruct | 7.65 B | not sold per token | 19.1 GB | ||
| Phi-tiny-MoE-instruct | 3.76 B | not sold per token | 9.4 GB |