48 GB VRAM and 864 GB/s of memory bandwidth. The strongest single-device fits at 4-bit are Gemma 4 31B (~55 tok/s), Qwen3 32B (~53 tok/s), QwQ 32B (~53 tok/s), gpt-oss-20b (~411 tok/s) and Gemma 3 27B (~63 tok/s).
36 open-weight models fit in L40S 48 GB at 4-bit, ranked by Analyzer Index. Renting costs $730 a month at 24 h/day.
| Model | Index | Memory | Est. tok/s | API $/month | Verdict |
|---|---|---|---|---|---|
| Gemma 4 31B Google31.3B | 57.4 | 19 GB | ~55.2 | $18.50 | API is cheaper |
| Qwen3 32B Alibaba (Qwen)32.8B | 50.0 | 20 GB | ~52.7 | $15.00 | API is cheaper |
| QwQ 32B Alibaba (Qwen)32.8B | 37.2 | 20 GB | ~52.7 | — | no API price |
| gpt-oss-20b OpenAI20.9B · 4.2B active | 35.2 | 13 GB | ~411 | $4.05 | API is cheaper |
| Gemma 3 27B Google27.4B | 28.5 | 16 GB | ~63.1 | $12.00 | API is cheaper |
| Llama 3.1 70B Instruct Meta70.6B | 24.3 | 42 GB | ~24.5 | $50.00 | API is cheaper |
| Llama 3.3 70B Instruct Meta70.6B | 20.9 | 42 GB | ~24.5 | $18.00 | API is cheaper |
| Nemotron 3.5 Lightning NVIDIA31.6B · 3B active | — | 19 GB | ~576 | $5.50 | API is cheaper |
| Qwen3.6 35B A3B Alibaba (Qwen)36B · 3B active | — | 22 GB | ~576 | $22.50 | API is cheaper |
| Qwen3.5-35B-A3B Alibaba (Qwen)36B · 3B active | — | 22 GB | ~576 | $25.25 | API is cheaper |
| Nemotron 3 Nano 30B A3B NVIDIA31.6B · 3B active | — | 19 GB | ~576 | $10.00 | API is cheaper |
| Qwen3 VL 30B A3B Thinking Alibaba (Qwen)31.1B · 3B active | — | 19 GB | ~576 | $54.00 | API is cheaper |
| Qwen3 VL 30B A3B Instruct Alibaba (Qwen)31.1B · 3B active | — | 19 GB | ~576 | $30.00 | API is cheaper |
| Qwen3 Coder 30B A3B Instruct Alibaba (Qwen)30.5B · 3B active | — | 18 GB | ~576 | $14.00 | API is cheaper |
| Qwen3 30B A3B Instruct 2507 Alibaba (Qwen)30.5B · 3B active | — | 18 GB | ~576 | $16.50 | API is cheaper |
| Qwen3 30B A3B Alibaba (Qwen)30.5B · 3B active | — | 18 GB | ~576 | $24.50 | API is cheaper |
| Ministral 3 3B 2512 Mistral3.8B | — | 2 GB | ~455 | $12.50 | API is cheaper |
| Gemma 4 26B A4B Google25.8B · 4B active | — | 16 GB | ~432 | $9.70 | API is cheaper |
| Gemma 3 4B Google4.3B | — | 3 GB | ~402 | $5.25 | API is cheaper |
| GLM 4.7 Flash Zhipu AI (GLM)31.2B · 4.6B active | — | 19 GB | ~376 | $16.05 | API is cheaper |
| Stable Diffusion 3.5 Large Stability AI8.1B | — | 5 GB | ~213 | — | no API price |
| Qwen3 VL 8B Instruct Alibaba (Qwen)8.8B | — | 5 GB | ~196 | $43.75 | API is cheaper |
| Ministral 3 8B 2512 Mistral8.9B | — | 5 GB | ~194 | $18.75 | API is cheaper |
| Qwen3.5-9B Alibaba (Qwen)9.7B | — | 6 GB | ~178 | $11.25 | API is cheaper |
| FLUX.1 [dev] Black Forest Labs11.9B | — | 7 GB | ~145 | — | no API price |
| Gemma 3 12B Google12.2B | — | 7 GB | ~142 | $7.00 | API is cheaper |
| Ministral 3 14B 2512 Mistral13.9B | — | 8 GB | ~124 | $25.00 | API is cheaper |
| Qwen3 14B Alibaba (Qwen)14.8B | — | 9 GB | ~117 | $15.50 | API is cheaper |
| Mistral Small 3 Mistral23.6B | — | 14 GB | ~73.2 | $7.00 | API is cheaper |
| Mistral Small 3.2 24B Mistral24B | — | 14 GB | ~72.0 | $12.50 | API is cheaper |
| Mistral Small 3.1 24B Mistral24B | — | 14 GB | ~72.0 | $48.97 | API is cheaper |
| Voxtral Small 24B 2507 Mistral24.3B | — | 15 GB | ~71.1 | $17.50 | API is cheaper |
| Qwen3.8 27B Alibaba (Qwen)27.8B | — | 17 GB | ~62.2 | $14.50 | API is cheaper |
| Qwen3.6 27B Alibaba (Qwen)27.8B | — | 17 GB | ~62.2 | $70.00 | API is cheaper |
| Qwen3.5-27B Alibaba (Qwen)27.8B | — | 17 GB | ~62.2 | $58.50 | API is cheaper |
| Muse Glimmer 30B Meta29.8B | — | 18 GB | ~58.0 | $60.00 | API is cheaper |
Memory need is weights × 1.2 for runtime overhead. Speed is a theoretical single-stream ceiling (memory bandwidth ÷ active-weight bytes); real engines usually reach 50–80% of it, and batching many requests raises total throughput well beyond one stream. Rental and purchase prices are typical figures checked 2026-10-14; power and engineering time are not included.