DeepSeek V4 Flash 0731 (batch)
DeepSeek · Released Jul 2026
A sparse mixture-of-experts model from DeepSeek with 284 billion total parameters and 13 billion active parameters, optimized for batch processing with a 1-million-token context window.
- Strengths
- Delivers frontier-class reasoning and text generation performance at a fraction of the compute cost due to its sparse architecture activating only a subset of parameters per request.
- Best for
- Batch jobs that process large document sets or logs where latency is not a constraint, allowing efficient use of the sparse architecture.
- Limitations
- Batch-only availability means it cannot serve real-time or streaming applications; superseded by V4.1 Flash released in September 2026.
Input / 1M
$0.11
Output / 1M
$0.33
Cached input / 1M
$0.0035
Context window
1.05M
Price history
Input (solid)Output (dashed)
Price change
- 30d
- in — out —
- 90d
- in decreased 21.4% out increased 17.9%
- 1y
- in decreased 21.4% out increased 17.9%
- Since launch
- in decreased 21.4% out increased 17.9%
Snapshots
| Effective | Input | Output | Cached in | Note | Source |
|---|---|---|---|---|---|
| 8 Sep 2026 | $0.11 | $0.33 | $0.0035 | Imported from OpenRouter | openrouter.ai |
| 29 Aug 2026 | $0.14 | $0.28 | $0.03 | Imported from OpenRouter | openrouter.ai |