# DeepSeek-V4-Flash

> OnPrem World · https://onpremworld.com/en/models/deepseek-v4-flash · Last updated: 2026-10-14 · CC BY 4.0

A 284B MoE with 13B active and a 1M-token context; at 4-bit it needs a 256 GB-class Mac. We can't model its compressed sparse attention yet, so our memory figures leave out the KV cache.

|  |  |
|---|---|
| Developer | DeepSeek |
| Parameters | 284B (13B active, MoE) |
| Native context | 1,048,576 tokens |
| Licence | MIT |
| Commercial use | Yes |
| KV cache | Specialised compressed attention (not estimated yet) |
| Released | 2026-04 |

## Memory needed

Weights + KV cache + runtime overhead, in GiB. Lower quantization = less memory, slightly lower quality.

| Quantization | 4K | 8K | 32K |
|---|---|---|---|
| Q4_K_M | 168.9 GiB | 168.9 GiB | 168.9 GiB |
| Q5_K_M | 198.0 GiB | 198.0 GiB | 198.0 GiB |
| Q6_K | 229.3 GiB | 229.3 GiB | 229.3 GiB |
| Q8_0 | 295.6 GiB | 295.6 GiB | 295.6 GiB |
| FP16 | 555.9 GiB | 555.9 GiB | 555.9 GiB |

## Which hardware runs it (8K)

| Hardware | Memory | Best quant | tok/s (est.) |
|---|---|---|---|
| M5 Ultra | 512 GB | Q8_0 | 56 |
| M3 Ultra | 512 GB | Q8_0 | 39 |
| GeForce RTX 3060 (12 GB) | 12 GB | Doesn't fit | – |
| GeForce RTX 3090 | 24 GB | Doesn't fit | – |
| GeForce RTX 4060 Ti 16GB | 16 GB | Doesn't fit | – |
| GeForce RTX 4080 SUPER | 16 GB | Doesn't fit | – |
| GeForce RTX 4090 | 24 GB | Doesn't fit | – |
| GeForce RTX 5060 Ti 16GB | 16 GB | Doesn't fit | – |
| GeForce RTX 5070 | 12 GB | Doesn't fit | – |
| GeForce RTX 5070 Ti | 16 GB | Doesn't fit | – |
| GeForce RTX 5080 | 16 GB | Doesn't fit | – |
| GeForce RTX 5090 | 32 GB | Doesn't fit | – |
| Radeon RX 7900 XT | 20 GB | Doesn't fit | – |
| Radeon RX 7900 XTX | 24 GB | Doesn't fit | – |
| Radeon RX 9060 XT 16GB | 16 GB | Doesn't fit | – |
| Radeon RX 9070 XT | 16 GB | Doesn't fit | – |
| Arc B580 | 12 GB | Doesn't fit | – |
| RTX A6000 | 48 GB | Doesn't fit | – |
| RTX 6000 Ada Generation | 48 GB | Doesn't fit | – |
| RTX PRO 5000 72GB Blackwell | 72 GB | Doesn't fit | – |
| RTX PRO 6000 Blackwell Workstation Edition | 96 GB | Doesn't fit | – |
| Radeon AI PRO R9700 | 32 GB | Doesn't fit | – |
| Arc Pro B60 | 24 GB | Doesn't fit | – |
| Arc Pro B70 | 32 GB | Doesn't fit | – |
| M4 Pro | 64 GB | Doesn't fit | – |
| M4 Max (14-core CPU / 32-core GPU) | 36 GB | Doesn't fit | – |
| M4 Max (16-core CPU / 40-core GPU) | 128 GB | Doesn't fit | – |
| M5 | 32 GB | Doesn't fit | – |
| M5 Pro | 64 GB | Doesn't fit | – |
| M5 Max (18-core CPU / 32-core GPU) | 36 GB | Doesn't fit | – |
| M5 Max (18-core CPU / 40-core GPU) | 128 GB | Doesn't fit | – |
| M6 (16GB configuration, 153 GB/s) | 16 GB | Doesn't fit | – |
| M6 (24GB/32GB configurations, 170 GB/s) | 32 GB | Doesn't fit | – |
| DGX Spark | 128 GB | Doesn't fit | – |
| RTX Spark (N1X) laptops & compact desktops | 128 GB | Doesn't fit | – |
| Ryzen AI Max+ 395 (Strix Halo) systems | 128 GB | Doesn't fit | – |
| Ryzen AI Max+ PRO 495 (Gorgon Halo) systems | 192 GB | Doesn't fit | – |

## Frequently asked questions

### How much VRAM does DeepSeek-V4-Flash need?

With an 8K context, about 168.9 GiB at Q4_K_M and 295.6 GiB at Q8_0 (weights + KV cache + runtime overhead). Longer contexts need more memory for the KV cache.

### Can DeepSeek-V4-Flash run on an GeForce RTX 4090?

DeepSeek-V4-Flash does not fit on the GeForce RTX 4090 (24 GB), even at Q4_K_M with an 8K context.

### Can I run DeepSeek-V4-Flash on a Mac?

Yes. DeepSeek-V4-Flash fits on the M3 Ultra (256 GB) at Q4_K_M with an 8K context, using about 168.9 GiB, at an estimated 68 tokens/s.

### Can DeepSeek-V4-Flash be used commercially?

DeepSeek-V4-Flash is released under MIT. The licence allows commercial use (read its conditions).

## Sources

- https://api-docs.deepseek.com/news/news260424
- https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash
- https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/resolve/main/config.json