Cosmos models
7 Cosmos models from NVIDIA on Hugging Face, from 2.4B to 64.6B parameters, published by NVIDIA in BF16, with contexts from 125K to 256K tokens. The smallest official model, Cosmos-Reason2-2B, needs about 5.5 GB of VRAM at its published precision; the cheapest live fit is RTX 3070 at $0.088/hr.
Pick a size
One row per official Cosmos size: the VRAM it needs at each precision, the cheapest GPU that holds it today, and the KV cache for a 32K-token context.
| Model | Parameters | Native VRAM | FP8 VRAM | INT4 VRAM | Live GPU fit (native) | Est. $/hr | KV cache at 32K |
|---|---|---|---|---|---|---|---|
| Cosmos-Reason2-2B | 2.4B | 5.5 GB | 2.7 GB | 1.4 GB | RTX 3070 | $0.088/hr | not published for this architecture |
| Alpamayo2-Edge | 2.7B | 5.9 GB | 3.0 GB | 1.5 GB | RTX 3070 | $0.088/hr | not published for this architecture |
| Cosmos-Reason1-7B | 8.3B | 18.5 GB | 9.3 GB | 4.6 GB | RTX A5000 | $0.176/hr | 1.8 GB |
| Cosmos-Reason2-8B | 8.8B | 19.6 GB | 9.8 GB | 4.9 GB | RTX A5000 | $0.176/hr | not published for this architecture |
| Cosmos-H-Surgical | 15.2B | 33.9 GB | 17.0 GB | 8.5 GB | RTX A6000 | $0.363/hr | not published for this architecture |
| Cosmos3-Super-Text2Image-4Step | 64.0B | 143 GB | 71.5 GB | 35.8 GB | AMD MI300X | $2.63/hr | 8.0 GB |
| Cosmos3-Super-Image2Video | 64.6B | 144 GB | 72.2 GB | 36.1 GB | AMD MI300X | $2.63/hr | 8.0 GB |
VRAM is weights times a flat 1.2 overhead; the KV cache is a separate per-model figure at 16-bit, shown where the architecture is published. See the methodology.
Official models (7)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| Cosmos-Reason2-2B | 2.4B | BF16 | 5.5 GB | RTX 3070 | $0.088/hr |
| Alpamayo2-Edge | 2.7B | BF16 | 5.9 GB | RTX 3070 | $0.088/hr |
| Cosmos-Reason1-7B | 8.3B | BF16 | 18.5 GB | RTX A5000 | $0.176/hr |
| Cosmos-Reason2-8B | 8.8B | BF16 | 19.6 GB | RTX A5000 | $0.176/hr |
| Cosmos-H-Surgical | 15.2B | BF16 | 33.9 GB | RTX A6000 | $0.363/hr |
| Cosmos3-Super-Text2Image-4Step | 64.0B | BF16 | 143 GB | AMD MI300X | $2.63/hr |
| Cosmos3-Super-Image2Video | 64.6B | BF16 | 144 GB | AMD MI300X | $2.63/hr |
VRAM is for the precision the model is published in: the weights times a flat 1.2 overhead, with the KV cache not included. See the methodology. The fit shown is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does.
More from NVIDIA
- Parakeet (8 models)
- Nemotron 3 (19 models)
- Nemotron Nano (6 models)
- nvidia (2 models)
- Nemotron Labs (5 models)
- Nemotron Speech (1 models)
- Nemotron Parse (1 models)
- Nemotron-H (3 models)
- Nemotron Cascade (1 models)
- Canary (2 models)
- Nemotron VoiceChat (1 models)