DeepSeek V4 Flash is the efficiency play in the V4 family: 284B total parameters, 13B active per token, the same 1M-token default context window as V4 Pro, and the same MIT license. Released alongside Pro on April 24, 2026.
At $0.14 per million input tokens and $0.28 per million output, Flash is one of the cheapest frontier-class APIs on the market — an order of magnitude below most proprietary flagships. It is also the V4 model that is realistically self-hostable, which makes it the default recommendation for teams that want DeepSeek-class capability under their own roof.
V4 Flash specifications
| Vendor | DeepSeek (China) |
|---|---|
| Architecture | Mixture-of-Experts with hybrid sparse attention (CSA + HCA) |
| Total parameters | 284B |
| Active parameters | 13B |
| Context window | 1M tokens |
| Modalities | text |
| License | MIT — weights free to download, self-host, fine-tune, and use commercially |
| Release date | 2026-04-24 |
| Weights | Hugging Face · GitHub |
V4 Flash API pricing
| Direction | Price per 1M tokens |
|---|---|
| Input | $0.14 |
| Output | $0.28 |
Among the cheapest frontier-class APIs available as of July 2026. Cache hits billed at one tenth of standard input. Verified July 2026.
Running V4 Flash locally
The self-hostable member of the V4 family. At 284B total / 13B active, a 4-bit quantization fits in roughly 150–190GB of memory — think a multi-3090/4090 rig, one H100/H200 node, or a 192GB Mac Studio class machine — while the 13B active footprint keeps token throughput high.
What V4 Flash is best for
- High-volume production workloads where cost per token dominates
- Self-hosting the V4 generation on a single node
- Latency-sensitive applications (13B active parameters per token)
Frequently asked questions
How is V4 Flash different from V4 Pro?
Flash is 284B total / 13B active parameters versus Pro's 1.6T / 49B. Both share the 1M-token context window and MIT license. Pro is stronger on hard reasoning; Flash is ~12× cheaper on the API and small enough to self-host on a single node.
What does the DeepSeek V4 Flash API cost?
As of July 2026: $0.14 per million input tokens and $0.28 per million output tokens, with cache hits at one tenth of input pricing.
Can I run DeepSeek V4 Flash locally?
Yes — it is the practical local option in the V4 family. A 4-bit quantization needs roughly 150–190GB of memory, so plan for a multi-GPU workstation, a single H100/H200-class node, or a 192GB unified-memory Mac.