DeepSeek V4.1 Flash (1017)
DeepSeek's main model, updated October 17, 2026. Open weights, a 1M token context window and very low API prices, which halve again at off-peak hours.
- Made by
- DeepSeek
- Released
- Type
- Open weights
- Status
- Live
- How to get it
- DeepSeek app and API (model name deepseek-flash); weights free to download on Hugging Face and ModelScope
- Good for
- Low-cost API useCodingSelf-hosting
What it is
DeepSeek launched V4.1 Flash on September 10, 2026, and made it the model behind its main deepseek-flash API name. The October 17 update, called Flash-1017, scores higher on DeepSeek’s benchmarks, finishes tasks faster and gives shorter answers. The weights are public, so anyone can download and run it.
What it costs
In the API, peak prices are $0.30 per million input tokens and $1.20 per million output tokens. Off-peak hours cost half that, and cached input is a fraction of a cent. The context window is 1 million tokens, with answers up to 384K tokens long.
What it means for you
DeepSeek is one of the cheapest strong models you can call, which makes it popular for coding tools and bulk jobs. As with any service, check where your data is processed before sending anything sensitive. Running the open weights on your own hardware avoids that question entirely.
Use cases this model is good for
How to automate a boring computer task with an AI-written script (no coding experience needed)
You do the same file-renaming or copy-paste job every week and suspect a small script could do it for you.
Best pick: ClaudeHours saved
