DeepSeek V4 Flash 0731
DeepSeek · Released Jul 2026
DeepSeek V4 Flash 0731 is a sparse mixture-of-experts model with 13 billion active parameters out of 284 billion total, designed for efficient inference across coding, reasoning, and agent tasks.
- Strengths
- Achieves frontier-class performance on coding, reasoning, and multi-step agent workflows while maintaining low computational overhead through its sparse architecture.
- Best for
- Building agents, implementing code generation features, and reasoning-heavy applications where you need fast inference without the latency of full-scale models.
- Limitations
- A re-post-trained revision with less field validation than the main V4 Flash release; tradeoffs between sparse efficiency and reasoning depth may apply to highly complex reasoning tasks.
Input / 1M
$0.09
Output / 1M
$0.18
Cached input / 1M
$0.018
Context window
1.05M
Price history
Snapshots
| Effective | Input | Output | Cached in | Note | Source |
|---|---|---|---|---|---|
| 2 Aug 2026 | $0.09 | $0.18 | $0.018 | Imported from OpenRouter | openrouter.ai |