Ornith 1 models
18 Ornith 1 models on Hugging Face, from 4.3B to 403.4B parameters. At the precision each one is published in, the smallest needs about 9.7 GB of VRAM (Ornith-1.5-9B-INT4-W4A16-AutoRound, cheapest live fit: RTX 5060 Ti) and the largest about 902 GB (Ornith-1.5-397B). The cheapest way to run Ornith-1.5-9B-INT4-W4A16-AutoRound is $0.110/hr.
Ornith 1 models
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| Ornith-1.5-9B-INT4-W4A16-AutoRound | 4.3B | BF16 | 9.7 GB | RTX 5060 Ti | $0.110/hr |
| Ornith-1.5-9B-MLX | 9.0B | BF16 | 20.0 GB | RTX A5000 | $0.176/hr |
| Ornith-1.5-9B-OBLITERATED | 9.7B | BF16 | 21.6 GB | RTX A5000 | $0.176/hr |
| Ornith-1.5-9B | 9.7B | BF16 | 21.6 GB | RTX A5000 | $0.176/hr |
| Ornith-1.5-35B-A3B-MLX | 34.7B | BF16 | 77.5 GB | A100 | $1.31/hr |
| Ornith-1.0-35B-AEON-Ultimate-Uncensored-BF16 | 35.1B | BF16 | 78.5 GB | A100 | $1.31/hr |
| Ornith-1.0-35B-FP8 | 35.1B | F8_E4M3 | 39.2 GB | RTX 4090 | $0.441/hr |
| Ornith-1.0-35B-FP8 | 35.1B | F8_E4M3 | 39.2 GB | RTX 4090 | $0.441/hr |
| Ornith-1.0-35B-AWQ-FP8 | 35.1B | F8_E4M3 | 39.3 GB | RTX 4090 | $0.441/hr |
| Ornith-1.0-35B-FP8 | 35.1B | F8_E4M3 | 39.3 GB | RTX 4090 | $0.441/hr |
| Ornith-1.5-35B-A3B | 36.0B | BF16 | 80.4 GB | RTX PRO 6000 | $1.38/hr |
| Ornith-1.5-35B-A3B-FP8 | 36.0B | F8_E4M3 | 40.2 GB | RTX 4090 | $0.441/hr |
| Ornith-1.0-397B | 396.8B | BF16 | 887 GB | No live fit | – |
| Ornith-1.0-397B | 396.8B | BF16 | 887 GB | No live fit | – |
| Ornith-1.0-397B-FP8 | 396.8B | F8_E4M3 | 443 GB | RTX PRO 6000 × 5 | $6.88/hr |
| Ornith-1.0-397B-FP8 | 397.0B | F8_E4M3 | 444 GB | RTX PRO 6000 × 5 | $6.88/hr |
| Ornith-1.5-397B | 403.4B | BF16 | 902 GB | No live fit | – |
| Ornith-1.5-397B-FP8 | 403.4B | F8_E4M3 | 451 GB | RTX PRO 6000 × 5 | $6.88/hr |
VRAM is for the precision the model is published in, with the same overhead and an 8,192-token context assumed on every page; 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.
