# DeepSeek V4.1 Flash

> DeepSeek's fast V4.1 model, called deepseek-flash on DeepSeek's own API. On the independent index it scores 32 at its max setting for well under a dollar a task.

- Maker: DeepSeek (https://api-docs.deepseek.com)
- Type: Reasoning, open weights
- Artificial Analysis Intelligence Index: 32 (max setting)
- Context window: 1M
- Output speed: 179 tokens a second
- Scores checked 30 October 2026 at https://artificialanalysis.ai/leaderboards/models (listed there as "DeepSeek V4.1 Flash 1017")

## Where to use it through an API

- DeepSeek: model id `deepseek-flash`, latest, fast
- Ollama Cloud: model id `deepseek-v4.1-flash`
- Vercel AI Gateway: model id `deepseek/deepseek-v4.1-flash-1017`

## Use it with every other AI at once

MultiScreenAI (https://www.multiscreenai.com) puts DeepSeek V4.1 Flash on one screen next to ChatGPT, Claude, Gemini and 300+ models: one message goes to every panel, and a timed, blind test mode shows which answers best. Pro is free with a follow on X, and you use your own API keys.
AI agents and coding harnesses (Claude Code, Codex, Cursor, Gemini CLI) can ask DeepSeek V4.1 Flash and other models at once through the MCP server at https://www.multiscreenai.com/api/mcp/ai; setup: https://www.multiscreenai.com/mcp