Cheapest way to run each model
For each model: the smallest consumer GPU that fits at 4-bit with an 8K conversation, the cheapest GPU rental in our price data that fits, and a hosted API price where we have one. Rental prices checked 2026-11-09; API prices checked Nov 3 2026.
| Model | Needs (4-bit, 8K) | Smallest GPU | Cheapest rental | API $ in / out per 1M |
|---|---|---|---|---|
| Llama 3.1 8B Instruct | 6.0 GB | RTX 4060 | $0.06/hr | $0.02 / $0.05 |
| DeepSeek-R1-0528-Qwen3-8B | 6.2 GB | RTX 4060 | $0.06/hr | $0.06 / $0.09 |
| Gemma 4 12B | 7.8 GB | RTX 3060 | $0.06/hr | – |
| Qwen3 14B | 10.3 GB | RTX 3060 | $0.06/hr | – |
| gpt-oss-20b | 11.9 GB | Arc A770 16 GB | $0.16/hr | – |
| Devstral Small 2 24B | 15.7 GB | RTX 3090 | $0.16/hr | – |
| Gemma 4 26B A4B | 17.5 GB | RTX 3090 | $0.16/hr | $0.13 / $0.4 |
| Qwen3.8 27B | 18.0 GB | RTX 3090 | $0.16/hr | $0.42 / $3 |
| Gemma 4 31B | 21.5 GB | RTX 3090 | $0.16/hr | $0.14 / $0.4 |
| Qwen3.8-35B-A3B (Empero Distill) | 21.9 GB | RTX 5090 | $0.33/hr | – |
| Qwen3 32B | 21.9 GB | RTX 5090 | $0.33/hr | – |
| Qwen2.5-Coder 32B | 22.0 GB | RTX 5090 | $0.33/hr | – |
| Nanbeige4.3 33B | 22.5 GB | RTX 5090 | $0.33/hr | – |
| Llama 3.1 70B Instruct | 45.2 GB | – | $0.40/hr | – |
| Llama 3.3 70B Instruct | 45.2 GB | – | $0.40/hr | – |
| gpt-oss-120b | 63.1 GB | – | $0.40/hr | $0.05 / $0.25 |
| Ornith-1.5-122B | 79.9 GB | – | $1.69/hr | – |
| DeepSeek-V3.2 | 406.0 GB | – | $25.52/hr | – |
| Mistral Large 4 (Le Chonk) | 635.5 GB | – | $31.16/hr | – |
| Kimi K3 | 1508.9 GB | – | – | – |
Detail pages contain referral and affiliate links, marked on each link. Disclosure.