With 10 GB VRAM, assuming 16 GB system RAM for partial offload, these are the best open models that fit. The top pick is Mistral Nemo 12B at Q4_K_M (~8.7 GB, ~61 tok/s): ollama run mistral-nemo:12b-instruct-2407-q4_K_M.
Best for coding: Qwen 2.5 Coder 7B (qwen2.5-coder:7b-instruct-q8_0), Code Llama 7B (codellama:7b-instruct-q8_0), gpt-oss 20B (MoE, MXFP4) (gpt-oss:20b).
Runtimes available for this hardware: Ollama, LM Studio, llama.cpp, vLLM.
Open the interactive checker for RTX 3080 10GB| # | Model | Quant | Memory needed | Fit | Speed | Command |
|---|---|---|---|---|---|---|
| 1 | Mistral Nemo 12B | Q4_K_M | 8.7 GB | fits in fast memory | ~61 tok/s | ollama run mistral-nemo:12b-instruct-2407-q4_K_M |
| 2 | Gemma 3 12B | Q4_K_M | 9.4 GB | fits in fast memory | ~56 tok/s | ollama run gemma3:12b |
| 3 | Llama 3.2 Vision 11B | Q4_K_M | 8.9 GB | fits in fast memory | ~58 tok/s | ollama run llama3.2-vision:11b |
| 4 | MiniCPM-V 2.6 8B | Q5_K_M | 7.5 GB | fits in fast memory | ~70 tok/s | ollama run minicpm-v:8b-2.6-q5_K_M |
| 5 | DeepSeek R1 Distill Qwen 7B | Q8_0 | 9.1 GB | fits in fast memory | ~56 tok/s | ollama run deepseek-r1:7b-qwen-distill-q8_0 |
| 6 | Qwen 2.5 VL 7B | Q4_K_M | 7 GB | fits in fast memory | ~76 tok/s | ollama run qwen2.5vl:7b |
| 7 | Qwen3 8B | Q4_K_M | 6.2 GB | fits in fast memory | ~88 tok/s | ollama run qwen3:8b |
| 8 | DeepSeek R1 0528 Qwen3 8B | Q4_K_M | 6.2 GB | fits in fast memory | ~88 tok/s | ollama run deepseek-r1:8b |
| 9 | Gemma 3n E2B (on-device) | Q8_0 | 7.8 GB | fits in fast memory | ~66 tok/s | ollama run gemma3n:e2b-it-q8_0 |
| 10 | Gemma 3 4B | Q8_0 | 5.8 GB | fits in fast memory | ~91 tok/s | ollama run gemma3:4b-it-q8_0 |
| 11 | Qwen3 4B | Q8_0 | 5.2 GB | fits in fast memory | ~104 tok/s | ollama run qwen3:4b-q8_0 |
| 12 | tev1 4B (Together AI, decision) | Q8_0 | 5.4 GB | fits in fast memory | ~99 tok/s | ollama run tev1:4b-q8_0 |
Estimates, not benchmarks. Model sizes are ollama.com download sizes; memory needed adds headroom for ~8k tokens of context; speeds are rough figures from memory bandwidth. Data updated 2026-10-14.
A project by OnDevice Labs · llms.txt