Blog
DeepSeek V4.1 Flash Is Now in Cursor. Here's What Cursor Is Claiming
Cursor just added DeepSeek's cheap open-weight model and says every effort level sits on its cost frontier, at $0.60 to $0.98 per task. Here's what that means, what it doesn't, and how I'd try it.
Caleb Dorris 26 October 2026, 8:08 PM 5 min
The short version
A cheap open-weight model just landed in the editor I live in.
- What happened: on October 26, Cursor announced that DeepSeek V4.1 Flash is available. You turn it on in Cursor Settings > Models.
- Cursor’s claim: every effort level sits on the CursorBench cost frontier, from $0.60 to $0.98 per task. At max effort, Cursor’s chart shows 47.7% on CursorBench 4.0 at $0.98.
- My take: it’s Cursor’s own benchmark, so treat it as a strong hint, not a verdict. I’m giving it a trial on low-stakes work, not switching.
On October 26, Cursor posted on X that DeepSeek V4.1 Flash is now available in Cursor. That caught my eye, because I’ve spent this week writing about cheap and open-weight coding models: Kimi K3.1 Pro closing the gap and whether to self-host MiniMax M3.1 Flash. Both of those meant leaving my editor to try them. This one is just a setting.
What Cursor actually said
Here’s the post, word for word:
DeepSeek V4.1 Flash is now available in Cursor! Every effort level sits on the CursorBench cost frontier: from $0.60 to $0.98 per task, no model scores higher. Turn it on in Cursor Settings > Models.
The attached chart plots CursorBench 4.0 score against average cost per task. The model is labeled “DeepSeek V4.1 Flash 1017,” and the marked point (max effort) is 47.7% at $0.98.
“On the cost frontier” is a narrower claim than it sounds. It doesn’t mean DeepSeek V4.1 Flash is the best model in Cursor. On the same chart, the models that score higher also cost more per task. It means that within its price range, from $0.60 to $0.98, Cursor didn’t find anything that scores better. For a lot of everyday tasks, that’s exactly the question I care about.
Two caveats. CursorBench is Cursor’s own benchmark, and I haven’t found an independent run of it. And “per task” is Cursor’s average across its benchmark tasks, not a price quote for your work. A big refactor costs more than a one-line fix on any model.
What DeepSeek V4.1 Flash is
DeepSeek released V4.1 Flash on September 10 as the smallest model in its new architecture family. DeepSeek describes it as a 552B-parameter mixture-of-experts model with 8B active parameters for input and 16B for output. That’s a big part of why it’s cheap to run.
The version in Cursor’s chart is the October 17 update, V4.1-Flash-1017. DeepSeek’s change log says it finishes tasks faster with shorter outputs, and lists 77.0 on DeepSWE v1.1 (up from 74.2) and 42.3 on Terminal-Bench 4.0 (up from 31.2). Those are DeepSeek’s own numbers.
It’s also open-weight: the 1017 weights are on Hugging Face under the MIT license. You won’t run a 552B model on your laptop, but open weights are what keep the hosted price honest, the same point I made about MiniMax.
Where it fits next to Kimi and MiniMax
In Moonshot’s Terminal-Bench 4.0 chart that I covered in the Kimi K3.1 Pro post, DeepSeek V4.1 Flash was the previous open-weight leader at 42.3%. Kimi K3.1 Pro claims a big jump past it. And in MiniMax’s own comparison from the MiniMax post, DeepSeek’s 1017 update had the higher DeepSWE score.
So DeepSeek isn’t the newest headline. What’s different is the distribution: it’s the one you can try today with a toggle inside Cursor, measured on Cursor’s benchmark by Cursor.
How I’d try it
My AI coding setup for October hasn’t changed: I use Grok 4.7 for almost everything. My rule for new models is a trial, not a switch. So here’s the plan:
- Turn it on in Cursor Settings > Models.
- Give it well-defined, low-stakes tasks first: write tests for one function, explain a legacy PHP include, tidy a small WordPress plugin. It’s the same short list I planned for MiniMax.
- Compare my own cost and cleanup time against what I’m using now, from my own usage, not the chart. Benchmarks are a starting point. My own repos are the real test.
- Check the data question before client code. Before any client repo touches a new model, I want to know who serves it and how the data is handled, under that client’s rules. I haven’t confirmed that for this model in Cursor yet, so it stays on my own projects until I have.
If you’re trying to figure out which AI tools make sense for your team, your site or your app, let’s talk.
FAQ
How do I turn on DeepSeek V4.1 Flash in Cursor?
Cursor says to enable it in Cursor Settings > Models.
How much does DeepSeek V4.1 Flash cost per task in Cursor?
Cursor’s chart puts its effort levels between $0.60 and $0.98 per task on average across CursorBench. That’s a benchmark average, not a quote. Real tasks vary, so check your own usage.
Is DeepSeek V4.1 Flash the best coding model in Cursor?
Not by Cursor’s chart. Higher-scoring models cost more per task. Cursor’s claim is that no model scores higher in the $0.60 to $0.98 range.
Is DeepSeek V4.1 Flash open source?
The weights are open. DeepSeek published V4.1-Flash-1017 on Hugging Face under the MIT license.
Sources
Need a web developer in Nashville?
PHP, WordPress and Next.js. Free quote, and I usually reply within a couple of hours.
caleb@calebdorris.com and call or text (615) 914-1073