← All hardware
Best LLMs for RTX 4070 Ti Super 16 GB
16 GB VRAM and 672 GB/s of memory bandwidth. The strongest single-device fits at 4-bit are gpt-oss-20b (~320 tok/s).
Loading…
16 GB VRAM and 672 GB/s of memory bandwidth. The strongest single-device fits at 4-bit are gpt-oss-20b (~320 tok/s).
14 open-weight models fit in RTX 4070 Ti Super 16 GB at 4-bit, ranked by Analyzer Index.
| Model | Index | Memory | Est. tok/s | API $/month | Verdict |
|---|---|---|---|---|---|
| gpt-oss-20b OpenAI20.9B · 4.2B active | 35.2 | 13 GB | ~320 | $4.05 | pays back in 198 mo |
| Ministral 3 3B 2512 Mistral3.8B | — | 2 GB | ~354 | $12.50 | pays back in 64 mo |
| Gemma 3 4B Google4.3B | — | 3 GB | ~313 | $5.25 | pays back in 152 mo |
| Stable Diffusion 3.5 Large Stability AI8.1B | — | 5 GB | ~166 | — | no API price |
| Qwen3 VL 8B Instruct Alibaba (Qwen)8.8B | — | 5 GB | ~153 | $43.75 | pays back in 18 mo |
| Ministral 3 8B 2512 Mistral8.9B | — | 5 GB | ~151 | $18.75 | pays back in 43 mo |
| Qwen3.5-9B Alibaba (Qwen)9.7B | — | 6 GB | ~139 | $11.25 | pays back in 71 mo |
| FLUX.1 [dev] Black Forest Labs11.9B | — | 7 GB | ~113 | — | no API price |
| Gemma 3 12B Google12.2B | — | 7 GB | ~110 | $7.00 | pays back in 114 mo |
| Ministral 3 14B 2512 Mistral13.9B | — | 8 GB | ~96.7 | $25.00 | pays back in 32 mo |
| Qwen3 14B Alibaba (Qwen)14.8B | — | 9 GB | ~90.8 | $15.50 | pays back in 52 mo |
| Mistral Small 3 Mistral23.6B | — | 14 GB | ~56.9 | $7.00 | pays back in 114 mo |
| Mistral Small 3.2 24B Mistral24B | — | 14 GB | ~56.0 | $12.50 | pays back in 64 mo |
| Mistral Small 3.1 24B Mistral24B | — | 14 GB | ~56.0 | $48.97 | pays back in 16 mo |
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.