Qwen · Apache 2.0
Strong general-purpose model with excellent coding and reasoning balance.
Qwen 2.5 14B Instruct (14.77B) requires 12.7 GB VRAM at recommended quality (Q6_K). At efficient quality (Q4_K_M), it fits in 8.5 GB VRAM, making it compatible with the NVIDIA GeForce RTX 4060 8GB. On NVIDIA GeForce RTX 4090, expect approximately 55 tok/s at Q5_K_M. For the best experience, Budget Home AI Server ($1,162) is recommended.
— OwnRig methodology, data updated 2026-03-15
| Quality | Quantization | VRAM | File Size |
|---|---|---|---|
| full | Q8_0 | 16.3 GB | 14.5 GB |
| recommended | Q6_K | 12.7 GB | 11 GB |
| recommended | Q5_K_M | 10.6 GB | 9 GB |
| efficient | Q4_K_M | 8.5 GB | 7.2 GB |
| compressed | Q3_K_M | 6.9 GB | 5.9 GB |
KV cache VRAM at Q6_K quality. Longer context = more memory.
| Context | KV Cache | Total VRAM |
|---|---|---|
| 2K | 205 MB | 12.9 GB |
| 4K | 512 MB | 13.2 GB |
| 8K | 1 GB | 13.7 GB |
| 16K | 1.9 GB | 14.6 GB |
| 32K | 3.8 GB | 16.5 GB |
Performance data for Qwen 2.5 14B Instruct across different hardware.
| Device | Quantization | Speed | Rating | Fits in VRAM |
|---|---|---|---|---|
| NVIDIA GeForce RTX 4060 Ti 16GB | Q4_K_M | 30 tok/s | Good | ✓ |
| NVIDIA GeForce RTX 4090 | Q5_K_M | 55 tok/s | Excellent | ✓ |
| Apple M4 Max (36GB Unified) | Q5_K_M | 38 tok/s | Good | ✓ |
| NVIDIA GeForce RTX 4060 8GB | Q3_K_M | 17 tok/s | Marginal | ✓ |
| NVIDIA GeForce RTX 4070 Ti 12GB | Q4_K_M | 30 tok/s | Good | ✓ |
| NVIDIA GeForce RTX 3080 10GB | Q3_K_M | 24 tok/s | Acceptable | ✓ |
| Apple M3 Pro (18GB Unified) | Q3_K_M | 5 tok/s | Marginal | ✓ |
Qwen 2.5 14B Instruct is commonly used with Cursor, Continue, Aider, Open WebUI, LM Studio. For an AI coding workflow, pair it with an embedding model like nomic-embed-text for local RAG.
Data confidence: estimated. Last updated: 2026-03-15. Source