Deepseek V4 Flash 0731 - Model Request

Feature request for product/service

AI Models

Describe the request

It seems the Cursor team is open to adding open-source models to the pool. If I’m not mistaken, you added the Kimi K3 in 1-2 days or less. The Deepseek V4 Flash 0731 is a gamechanger for the current model distribution. I’d love to see it in the current model pool. Yes, there are alternatives, but they’re not that cheap.

Operating System (if it applies)

Windows 10/11

Created an account to say I would also greatly appreciate support for dsv4f 0731 being added.

Hi all, thank you for weighing in! We are tracking this feature request, and we’ll let you know any updates

Not a paid customer, but I would for sure if I had ds v4 flash 0731, for what it’s worth.

Likely many others in my same boat.

Even though the idea is very good and it wouldn’t be bad to have it on the list of models; since it is quite good and very economical, although, to be honest, it is below in coding quality compared to Composer 2.5, but if we have moonshot, why not have this one from deepseek.

Created account just to show support on this model request.
Deepseek V4 Flash 0731 is better at lower cost than the current least expensive model in Cursor. This would be best for simple tasks.

Deepseek V4 Pro 0813

pro version would be great too

Kimi K3 was released so immediately. Hoping deepseek v4 flash 0731 gets added too!

For now GPT 5.6 Luna is a reasonable replacement for deepseek v4 flash… Can’t really know how subsidized / precarious it is since its a ClosedAI model, but per the benchmarks seems legit ( agentic benchmark and avg cost per task )

Will take a look and maybe report back… this is from employee of one of cursor’s large enterprise customers

Same as above, I hope the Deepseek Flash or Pro model can appear in the official model list!

commenting for reach! open source models are good enough and cursor including it on the subs would be insane value for third party bucket

here to join the choir, add deepseek back!1

+1 for adding DeepSeek-V4.1-Flash (deepseek-flash) to the official model pool. This thread asked for V4 Flash 0731; V4.1-Flash is the current Flash model and would be a great cheap option for lighter agent work.

+1,A harness without DeepSeek-4.1-Flash is closed and arrogant, completely disregarding users’ feelings.

Please add DeepSeek V4.1 Flash ASAP

Please consider adding DeepSeek V4.1 Flash to Cursor as soon as possible.

This is not just another cheap “Flash” model. It offers an unusually strong combination of speed, intelligence, coding/agent capability, and price — without the typical “fast = much less intelligent” trade-off.

Why V4.1 Flash?

Metric DeepSeek V4.1 Flash
Artificial Analysis Intelligence Index 40
Artificial Analysis output speed ~210 tokens/s
Time to First Token (TTFT) ~1.37s
Terminal-Bench 2.1 90.6
DeepSWE v1.1 74.2
Codeforces Rating 3471
Context Window 1M tokens
API Input $0.30 / 1M
API Output $1.20 / 1M
Off-peak Input $0.15 / 1M
Off-peak Output $0.60 / 1M

Source: Artificial Analysis + DeepSeek official API/benchmark data.

1. Extremely fast, without the usual intelligence downgrade

Artificial Analysis currently measures V4.1 Flash at approximately 210 output tokens/sec, while giving it an Intelligence Index score of 40.

DeepSeek itself states that V4.1 Flash has surpassed V4 Pro in performance, cost, speed, and total runtime.

This is exactly what a coding agent needs: Flash-level latency without paying a huge “intelligence tax.”

2. Very strong coding & agent performance

This isn’t just a fast chat model:

  • 90.6 — Terminal-Bench 2.1
  • 74.2 — DeepSWE v1.1
  • 65.4 — NL2Repo-Bench
  • 3471 — Codeforces Rating
  • 90.9 — GPQA Diamond

These are particularly relevant to Cursor because repository understanding, terminal interaction, tool use, and multi-step software engineering are exactly where an agent model needs to perform.

3. The price/performance is extremely compelling

Model Input / 1M Output / 1M
DeepSeek V4.1 Flash (off-peak) $0.15 $0.60
DeepSeek V4.1 Flash (peak) $0.30 $1.20
Composer 2.5 $0.50 $2.50
Grok 4.6 $2.00 $6.00
Composer 2.5 Fast $3.00 $15.00

At off-peak pricing, V4.1 Flash output is roughly:

  • 10× cheaper than Grok 4.6
  • 4.2× cheaper than Composer 2.5
  • 25× cheaper than Composer 2.5 Fast

For agentic coding involving hundreds of model/tool calls and repeated edits, this difference is enormous.

4. Developer adoption is already happening

The ecosystem is already moving quickly.

OpenCode is an official V4.1 Flash launch partner, and its public usage data currently reports approximately 281K unique users and 4.59M completed sessions for V4.1 Flash, ranking it #1 in recent OpenCode usage.

Other coding-agent ecosystems, including CommandCode, are also moving quickly around low-cost, high-performance models like V4.1 Flash.

This demonstrates real developer demand for exactly this combination: fast + intelligent + cheap.

Cursor should not be late to this model.

Request

Please add DeepSeek V4.1 Flash to Cursor’s model selector as soon as possible.

For high-frequency coding and agent workflows, it has the potential to be one of the best price/performance models available today — and importantly, its speed does not appear to come from simply making the model dramatically less capable.

+1, but don’t get your hopes up. We have threads for adding Deepseek v4 Flash 0731, GLM-5.3, GLM-5.3 Flash, and even Composer 3 that have all gone varying amounts of time without being added - or even getting a (proper) response.

The first time we’ll likely get real news on any of these is after Grok 4.7 drops, along with whatever “small model” replacement we get for Composer 2.5 - whether that’s Composer 3 or Grok “Lite”.

I recommend starting to look elsewhere at other agentic development platforms that allow easier access to new models. We have to show Anysphere and SpaceXAI that we’re serious about leaving if they don’t start addressing these needs. I’ve already been experimenting with T3 Code + OpenCode, others have gone back to Codex and Claude Code, and some others are exploring options like OpenHands.

EDIT: post got merged, but I’m leaving my actual reply from the pre-merge thread as-is

+1 for adding DeepSeek-V4.1-Flash