What builders are saying about Grok 4.7 in Cursor

xAI released Grok 4.7 on September 21. The developer guide lists it on all Cursor plans, and Cursor's pricing docs put it in the Cursor Models pool at $2 input and $6 output per million tokens. We read public posts on X from September 23 to October 14, 2026 to see how people using it in Cursor describe it.

This is a sample, not a survey. We searched X for posts that mention Grok 4.7, mostly in Cursor and Grok Build. We left out automated accounts, bot replies, and promotional posts, and grouped first-person reports by theme. A theme here means at least two independent accounts describe the same thing. The quotes are copied verbatim, typos included. Dates are Pacific Time. We did not verify any of the claims in them.

Theme 1: Fast, and the allowance lasts

The most common positive report is about throughput and the meter, not cleverness.

"Grok 4.7 is great for me. Huge workhorse, gets so much done, my usage meter barely moves. Speed seems to depend on the platform. Cursor cloud agent grok 4.7 feels slow. Grok Build is super fast. Regular Grok 4.7 in cursor is fast."

@WR4NYGov, October 5

"I've been using Grok 4.7 on Cursor and surprisingly, it's better than Sonnet 5 for coding. Frontend, backend, debugging and working across the codebase, the overall performance has been really good."

@santhosh_patell, September 29

The counterpoint is that "fast" depends on the comparison. One developer ran the same one-shot build through two tools:

"One prompt, one try: build a mechanical pocket watch. Claude Code, Opus 5.5 - 5min 51s ($1.18) Cursor, Grok 4.7 - 12min 39s ($1.14) Both passed the same four checks. Grok saved 4 cents and took 2.2× as long."

@itworkstho, October 6

Theme 2: A good second model, not the only one

Several builders describe giving Grok 4.7 specific jobs (orchestration, search, small edits, bug hunting) and sending the hardest step to Claude.

"Grok 4.7 has been great for orchestration, especially helping me with my AWS task, and then composer and grok 4.7 is still pretty good for small coding tasks. For the heavy coding it’s either sonnet 5.5 or Opus 5.5…"

@baguley61, October 5

"Opus 5.5 still writes the cleanest patches for me. Grok 4.7 still finds the weird bugs first."

@ReaperGhost187, October 11

"I'd let Grok 4.7 do the volume work. Searching the code, writing tests, the boring boilerplate. The one step that needs real thinking goes to a stronger model. And the context stays small unless the task really needs the whole repo."

@Damir_Akaza, October 5

Theme 3: Frontend and visual design are the weak spot

This is the most specific negative theme, and it appears in independent tests.

"We ran it through BridgeBench's lava lamp test against Grok 4.6. Same prompt. Grok 4.6: $0.23, 9m 54s Grok 4.7: $0.35, 15m 46s … And the output is still not usable for real frontend work."

@bridgebench, September 22. This is a benchmark account, not an individual user.

"Floating island with pagoda in three.js. Bad compared to DeepSeek-v4.1-flash result i got. … Overall quality is off, appears like a very lazy and approximate implementation."

@ItsmeAjayKV, September 21

"I just spent weeks trying to coax a professional TailWindcss-style site out of grok and cursor - comical. Then I … had DDGai create a prompt for Claude Opus 5.5 in Cursor and 2 hours later it produced an incredible 7-site page family."

@MusicArtFilm, October 10

Theme 4: Runs long, and loses to Opus 5.5 on hard work

Most negative posts compare Grok 4.7 directly to Opus 5.5, and several describe extra tokens and tool calls rather than outright failure.

"I’ve been pretty disappointed with Grok 4.7 and it honestly feels like 4.6 that just runs longer and overcomplicates things with excessive tokens and as many tool calls at it can at a task."

@brennen_harrisX, September 25

"Went back to use grok 4.7 (cursor api usage is gone) and after being in a honeymoon with Opus 5.5….. wow grok 4.7 is so so bad."

@PedroAnibarro, October 12

"I was very bullish on Grok during 4.6 and thought 4.7 would be worth the wait. Anthropic and OpenAI is just too far ahead. I have to cancel my cursor subscription."

@johkyp, September 25

A milder version of the same view:

"Grok 4.7 brings general improvements over 4.6 with unchanged pricing, though I haven't seen any standout leaps."

@StashTrevor, September 28

Theme 5: Confusion about which pool pays

The third group of posts is about billing, not the model. It got louder after The X Subscription moved Claude and GPT in Cursor to API rates.

"Grok 4.7 is listed in the @cursor_ai app as a cursor model - but with only 4% USED out of my allocation, I was just told that I cannot use Grok 4.7 anymore and I need to switch to a cursor model."

@johnaspinallai, September 26

"They don't even told anyone that grok 4.7 was automatically user in cursor cloud agents with the huge surprise that grok 4.7 usage is not even included in your credits: you pay for that unless you tell the bot to use the previous model…"

@Daniele_molo17, October 7

"Using Claude and other models in the new Cursor pricing is billed separately at API rate and not included in the plan."

@iannuttall, October 13

We did not confirm either of the first two accounts. Here is what the published pages say:

• Cursor's docs list Grok 4.7 in the Cursor Models pool on legacy Pro, Pro Plus, and Ultra. Long context above 256K input tokens is billed at 2x, and Fast mode at 2x.
• x.ai/pricing says The X Subscription has one weekly pool shared across Grok and Cursor. Third-party models in Cursor are billed separately at API list price.
• The developer guide says Grok 4.7 Fast runs only in Cursor and Grok Build, billed through your plan.

Where Grok 4.7 stands

Putting the themes next to the published numbers:

• Strong: price and allowance. Grok 4.7's $6 output price is the lowest of the flagship models we compared (API comparison). In Cursor it sits in the included pool. Builders consistently report the meter moving slowly. Speed in Grok Build and Cursor's regular agent gets positive reports. Several users find it useful for orchestration, search, tests, and bug hunting next to a stronger model.
• Weak: frontend and visual design, which three independent reports call out. Long agent runs with many tool calls and tokens. Head-to-head quality against Opus 5.5, where most of the negative posts land. One heavy user reported Cursor cloud agents as slower than the local agent.
• Unclear: whether Grok 4.7 beats Sonnet 5.5. We found claims on both sides and no controlled comparison. Long context is cheap per token but adds up fast: one user's example works out to $0.80 a call at 400K tokens, and the API rate doubles above 200K.

The pattern

Across this sample, the people happiest with Grok 4.7 use it as a high-volume model, with Claude kept for the hardest step. The least happy are those who expected it to replace Opus 5.5, especially for frontend work. Billing posts are a separate problem from model quality, and the vendor pages above answer most of them. This is a sample of people who chose to post. It cannot tell you which group is larger.

Background: Grok 4.7 launch coverage, Grok and Cursor are now one bill, and Before you switch to The X Subscription.