DeepSeek Coder models
6 DeepSeek Coder models from DeepSeek on Hugging Face, from 1.3B to 15.7B parameters, published by DeepSeek in BF16, with contexts from 4K to 160K tokens. The smallest official model, deepseek-coder-1.3b-instruct, needs about 3.0 GB of VRAM at its published precision; the cheapest live fit is RTX 3070 at $0.088/hr.
Part of the DeepSeek series
Pick a size
One row per official DeepSeek Coder size: the VRAM it needs at each precision, the cheapest GPU that holds it today, and the KV cache for a 32K-token context.
| Model | Parameters | Native VRAM | FP8 VRAM | INT4 VRAM | Live GPU fit (native) | Est. $/hr | KV cache at 32K |
|---|---|---|---|---|---|---|---|
| deepseek-coder-1.3b-instruct | 1.3B | 3.0 GB | 1.5 GB | 0.8 GB | RTX 3070 | $0.088/hr | 6.0 GB |
| deepseek-coder-6.7b-instruct | 6.7B | 15.1 GB | 7.5 GB | 3.8 GB | RTX A4000 | $0.167/hr | 16.0 GB |
| deepseek-coder-7b-instruct-v1.5 | 6.9B | 15.4 GB | 7.7 GB | 3.9 GB | RTX A4000 | $0.167/hr | 15.0 GB |
| DeepSeek-Coder-V2-Lite-Instruct | 15.7B | 35.1 GB | 17.6 GB | 8.8 GB | RTX A6000 | $0.363/hr | 0.95 GB |
VRAM is weights times a flat 1.2 overhead; the KV cache is a separate per-model figure at 16-bit, shown where the architecture is published. See the methodology.
Official models (5)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| deepseek-coder-1.3b-instruct | 1.3B | BF16 | 3.0 GB | RTX 3070 | $0.088/hr |
| deepseek-coder-6.7b-base | 6.7B | BF16 | 15.1 GB | RTX A4000 | $0.167/hr |
| deepseek-coder-6.7b-instruct | 6.7B | BF16 | 15.1 GB | RTX A4000 | $0.167/hr |
| deepseek-coder-7b-instruct-v1.5 | 6.9B | BF16 | 15.4 GB | RTX A4000 | $0.167/hr |
| DeepSeek-Coder-V2-Lite-Instruct | 15.7B | BF16 | 35.1 GB | RTX A6000 | $0.363/hr |
Quantized versions (1)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| DeepSeek-Coder-V2-Lite-Instruct-FP8 | 15.7B | F8_E4M3 | 17.6 GB | RTX 4000 Ada | $0.220/hr |
VRAM is for the precision the model is published in: the weights times a flat 1.2 overhead, with the KV cache not included. See the methodology. The fit shown is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does.
More from DeepSeek
- DeepSeek R1 (14 models)
- DeepSeek V3 (7 models)
- DeepSeek OCR (2 models)
- DeepSeek V2 (5 models)
- DeepSeek VL (3 models)
- DeepSeek MoE (2 models)
- DeepSeek V4 (3 models)