DeepSeek Flash
The workhorse model. Fast, cheap, multimodal.
Input (off-peak)
$0.15
Output (off-peak)
$0.60
Context
1M
Concurrency
2500
Specifications
| Property | Value |
|---|---|
| Model id | deepseek-v4.1-flash-1017 |
| Context window | 1,000,000 tokens |
| Max output | 384,000 tokens |
| Vision input | Yes |
| Max concurrency | 2500 |
| Input, cache hit (off-peak) | $0.003 / 1M |
| Input, cache miss (off-peak) | $0.15 / 1M |
| Output (off-peak) | $0.60 / 1M |
| Peak multiplier | 2.0x |
What it is good at
- Classification and routing
- Structured extraction (JSON from documents)
- Translation and rewriting
- Summarisation
- Screenshot and document OCR (vision)
- Everyday chat
What it is weaker at
Long chains of reasoning where every step depends on the previous one: multi-file code review, financial analysis, contract interpretation. Flash tends to stop at the first plausible answer. That is exactly what V4 Pro is for.
Cheaper than official
Flash's open weights are resold by 45 providers, ranging from $0.025 to $0.45 per million input tokens — an 18x spread. Most of the cheap quotes use FP4 quantization. See provider comparison before switching.