SOLAR models
6 SOLAR models from upstage on Hugging Face, from 10.7B to 250.3B parameters, published by upstage in F16 and BF16, with contexts from 4K to 1M tokens. The smallest official model, SOLAR-10.7B-Instruct-v1.0, needs about 24.0 GB of VRAM at its published precision; the cheapest live fit is RTX A5000 at $0.176/hr.
Pick a size
One row per official SOLAR 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 |
|---|---|---|---|---|---|---|---|
| SOLAR-10.7B-Instruct-v1.0 | 10.7B | 24.0 GB | 12.0 GB | 6.0 GB | RTX A5000 | $0.176/hr | 6.0 GB |
| solar-pro-preview-instruct | 22.1B | 49.5 GB | 24.7 GB | 12.4 GB | A100 | $1.31/hr | not published for this architecture |
| Solar-Open-100B | 102.7B | 229 GB | 115 GB | 57.4 GB | RTX A6000 × 5 | $1.82/hr | 6.0 GB |
| Solar-Open2-250B | 250.3B | 559 GB | 280 GB | 140 GB | AMD MI300X × 3 | $7.89/hr | 6.0 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 (5)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| SOLAR-10.7B-Instruct-v1.0 | 10.7B | F16 | 24.0 GB | RTX A5000 | $0.176/hr |
| SOLAR-10.7B-v1.0 | 10.7B | F16 | 24.0 GB | RTX A5000 | $0.176/hr |
| solar-pro-preview-instruct | 22.1B | BF16 | 49.5 GB | A100 | $1.31/hr |
| Solar-Open-100B | 102.7B | BF16 | 229 GB | RTX A6000 × 5 | $1.82/hr |
| Solar-Open2-250B | 250.3B | BF16 | 559 GB | AMD MI300X × 3 | $7.89/hr |
Fine-tunes and community models (1)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| YanoljaNEXT-EEVE-Instruct-10.8B | 10.8B | BF16 | 24.2 GB | RTX A6000 | $0.363/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.