DeepSeek V4.1 Flash lands in Cursor: up to 47.7% on CursorBench for under $1 a task
Cursor added DeepSeek V4.1 Flash, a DeepSeek coding model with several effort levels. Cursor says every effort level sits on its CursorBench 4.0 cost frontier, at $0.60 to $0.98 per task, with no model scoring higher at those prices. Cursor's chart puts the top (max) level at 47.7% for $0.98 a task.
Why it matters
A cheap model that holds its own on Cursor's own benchmark is a good default for routine agent work, so you save pricier models for the hard tasks.
Who it's for
Cursor users who want to cut agent costs on everyday coding tasks, and anyone curious about open-weight models.
How to try it
- Open Cursor Settings > Models and turn on DeepSeek V4.1 Flash.
- Try a lower effort level for quick edits and max for harder tasks.
- Run the same task with your usual model and compare the result and cost before switching your default.
Plan & availability: Available now in Cursor. Turn it on in Cursor Settings > Models.
Source: @cursor_ai on X ↗
Written by AgentStackList from the original post. Checked Oct 27, 2026. Features, plans and availability can change, so confirm on the source before you rely on it.
cursordeepseekopen-weightmodel-pricingcursorbench
More Cursor tricks
- A fake Claude Code extension tried to plant an SSH backdoor on up to 5,000 Cursor users. Run this one-line check · Oct 29
- Cursor now lets you draw on your app to tell the agent what to fix: meet Annotate in design mode · Oct 27
- Cursor's Agent SDK is out: build a PR-fixing or Slack bot as a folder of code and deploy it with one command · Oct 26
- Cursor's agents can now click through your desktop apps to test what they build · Oct 23
Agent Stack Weekly · free
Get the week's best AI tricks
Limit and pricing changes, new MCP servers, and setup fixes for Claude Code, Codex, Cursor, Copilot and Antigravity.
No spam. Unsubscribe anytime. Read past issues