DeepSeek model prices across providers
15 models, 197 quotes. Sorted by how many providers offer each model.
| Model | Providers | Input range | Spread |
|---|---|---|---|
| DeepSeek V4.1 Flash 1017 | 45 | $0.025 – $0.450 | 18.0x |
| DeepSeek V4.1 Flash | 40 | $0.033 – $0.450 | 13.6x |
| DeepSeek V4 Flash 0731 | 29 | $0.059 – $0.440 | 7.5x |
| DeepSeek V4 Pro 0813 | 25 | $0.103 – $1.65 | 15.9x |
| DeepSeek V4 Flash 0423 | 17 | $0.050 – $0.600 | 12.0x |
| DeepSeek V4 Pro 0423 | 16 | $0.880 – $1.91 | 2.2x |
| DeepSeek V3.2 | 7 | $0.209 – $1.00 | 4.8x |
| DeepSeek V3.1 | 4 | $0.250 – $0.650 | 2.6x |
| R1 0528 | 3 | $0.500 – $0.571 | 1.1x |
| DeepSeek V3 | 3 | $0.256 – $0.320 | 1.2x |
| DeepSeek V4.1 Flash (batch) | 2 | $0.050 – $0.112 | 2.2x |
| DeepSeek V4 Flash Vision Exp | 2 | $0.216 – $0.440 | 2.0x |
| DeepSeek V3 0324 | 2 | $0.250 – $0.290 | 1.2x |
| DeepSeek V3.2 Exp | 1 | $0.270 – $0.270 | 1.0x |
| DeepSeek V3.1 Terminus | 1 | $0.270 – $0.270 | 1.0x |
USD per million input tokens, verified 2026-11-04. Source: OpenRouter public API.
Why prices differ so much
DeepSeek publishes open weights, so anyone with GPUs can serve the same model. Providers differ in:
- Quantization — FP4 runs on cheaper hardware than FP16, and shows in the price
- Hardware generation — newer GPUs cost less per token
- Utilisation — a full cluster is cheaper per token than a half-empty one
- Margin — some providers price for volume, some for margin
Machine-readable data
GET https://deepseekbase.com/data/providers.json