DeepSeek V4-Flash
V4-Flash brings the V4 architecture down to 284B total / 13B active parameters, keeping the 1M-token context while drastically reducing inference cost.
Key features
DeepSeek V4-Flash · AI Models
Parameters
284B total · 13B active
Context window
1M tokens
Released
April 24, 2026
Pricing
Lower-cost DeepSeek API tier
Key features
- 284B total / 13B active parameters.
- Same V4 hybrid attention as V4-Pro.
- 1M-token context window.
- Significantly cheaper than V4-Pro.
Best for
Use V4-Flash for high-throughput chat, classification and coding workloads where you want V4 quality at a lower price point.
Frequently Asked Questions
How does DeepSeek V4-Flash compare to V4-Pro?
V4-Flash uses 284B total / 13B active parameters vs V4-Pro's 1.6T / 49B. Flash is significantly cheaper and faster while sharing the same hybrid attention architecture.
Can I self-host DeepSeek V4-Flash?
DeepSeek makes V4-Flash available via their API. The open-weight model enables enterprise deployments that need cost-efficient V4 quality on their own infrastructure.
Open Chat
V4-Flash brings the V4 architecture down to 284B total / 13B active parameters, keeping the 1M-token context while drastically reducing inference cost.