DeepSeek V4.1 Flash 1017
October update of DeepSeek V4.1 Flash, tuned for whole-project agentic coding.
DeepSeek V4.1 Flash 1017 refreshes the September V4.1 Flash weights on the same CED mixture-of-experts architecture. DeepSeek says it scores higher while finishing tasks faster with shorter outputs, after extra reinforcement learning on end-to-end projects. It is the model behind the deepseek-flash API name.
- Provider
- DeepSeek
- Type
- Language model
- Released
- Oct 17, 2026
- Context window
- 1M tokens
- Price
- $0.30 input / $1.20 output per 1M tokens
- Input
- text, image
- Output
- text
- Open weights
- Yes
- License
- MIT
Best for
- High volume / low cost
- Real-time / low latency
- Coding
- Agents
- Local & on-device
Strengths
- Open weights (MIT)
- Shorter, faster task runs
- Low API price
More from DeepSeek
- DeepSeek V4.1 Flash — DeepSeek. Open-weight MoE built on DeepSeek’s new encoder-decoder architecture.
- DeepSeek V4 Pro — DeepSeek. DeepSeek’s large open MoE, GA release (0813).