
Grok vs ChatGPT: Grok 4.7 and GPT-6 Astra on the Same 3 Builds, Plus What Each Plan Costs
Pat Simmons's tests put Grok 4.7 against ChatGPT's GPT-6 Astra on three identical builds and Grok Bot against ChatGPT's Dots on three agent jobs, next to what every plan costs.
For building and coding, ChatGPT won. In Pat Simmons's Grok 4.7 vs Fable 5.1 vs GPT-6 Astra build costs test, ChatGPT's top model, GPT-6 Astra, beat Grok 4.7 on all three identical builds, and Grok 4.7 placed last in each one. Grok cost far less to run: $29.52 for all three builds, against $114.29 for Astra.
For agent jobs, Grok came out on top. In Pat's Grok Bot vs ChatGPT Dots vs Meta Muse on the same three jobs test, Grok Bot got an A- for running a fake coffee company for four hours, and ChatGPT's Dots got a B.
On subscriptions, ChatGPT has a free plan, and ChatGPT Plus ($20 a month) includes GPT-6 Astra. Grok's plans page lists four paid plans from $8 to $200 a month, and Grok Bot starts on the $30 Premium plan. Every plan price on this page was checked on the companies' own pages on November 5, 2026.
| Pat's 3 builds | Last in all 3 | Placed above Grok in all 3 |
| API price per million tokens (input / output) | $2 / $6 | $10 / $50 |
| Cost of all 3 builds | $29.52 | $114.29 |
| Cheapest plan that lists the top model | Premium, $30 a month ("Latest Grok models") | ChatGPT Plus, $20 a month |
| Agent, and the plan it needs | Grok Bot: A- in Pat's test, from Premium ($30) | Dots: B in Pat's test, with Pro (from $100) |
Is Grok better than ChatGPT?
Grok isn't better for building and coding: GPT-6 Astra, the top model in ChatGPT, beat Grok 4.7 in all three of Pat's builds. The jobs were a copy of an award-winning website, a 3D product page for a mechanical keyboard, and a playable copy of Mario Kart's Rainbow Road.
Grok wins on price per token and on agent work. Grok 4.7 costs $2 per million input tokens and $6 per million output tokens on the model page of SpaceXSI, the company that makes Grok, against $10 and $50 for GPT-6 Astra on OpenAI's API pricing page. Tokens are the small chunks of text a model reads and writes, and these are the prices developers pay to use each model in their own apps. And Grok Bot beat ChatGPT's Dots in Pat's three-job agent test.
Pat's verdict on Grok 4.7, from his review: "I was all ready to give 4.7 the benefit of the doubt, especially because it's so much cheaper, but it really was not coming anywhere close to Fable and Astra."
SuperGrok vs ChatGPT Plus: what each plan costs
ChatGPT Plus is $20 a month and includes GPT-6 Astra. Grok's closest match is the $30 Premium plan, which replaced SuperGrok for new sign-ups: SpaceXSI's X Subscription announcement says Premium includes "the benefits of SuperGrok, Cursor Pro, and X Premium+." The same announcement says existing SuperGrok subscribers can keep their current plan and price.
Both companies call one of their plans "Plus." ChatGPT Plus is the $20 plan, and Grok's Plus is the $8 entry plan. Here is every individual plan on each pricing page, checked November 5, 2026.
| Plan | Price | Top model on the plan | Agent | Also on the plan |
|---|---|---|---|---|
| $0 | GPT-6 Luna | None | Unlimited text chats with GPT-6, within OpenAI's abuse guardrails; limited uploads, images, voice and deep research | |
| $8 a month | GPT-6 Luna | None | More messages, uploads, images and voice than Free; the page says it "may include ads" | |
| $20 a month | GPT-6 Astra | None | Expanded deep research, Codex and ChatGPT Work | |
| From $100 a month | GPT-6 Astra, with more use | Dot, "your always-on agent" | Three usage tiers (Pro 100, Pro 200 and Pro 500) and up to $500 a month in API credits | |
| $8 a month | Not named on the page | None | Grok chat, voice, Imagine and Build, Cursor, and X perks like the checkmark | |
| $30 a month | "Latest Grok models" | Grok Bot | 4 times the usage of Plus, AI images and videos, cloud agents | |
| $100 a month | Everything in Premium | Grok Bot | 15 times the usage of Plus, priority access at busy times | |
| $200 a month | "Most powerful Chat models" | Grok Bot | 50 times the usage of Plus, earliest access to new products and models |
Grok's plans page describes one "shared AI usage pool across all products," so time spent in Grok Bot comes out of the same allowance as Grok chat and Cursor. On November 5 the plans page also showed a first-month offer of half off Premium, Super and Ultra, and it listed no free plan. Grok's X Subscription tiers explained covers what changed when the plans merged.


Pat paid for SuperGrok during his agent test and called it "$30 a month, not to be confused with Cursor Pro, which is $20 a month." He used it all up and added that "realistically you have to have at least that $100 subscription if you want to get real work done every day with Grok Bot." For the other chatbots and their prices, see ChatGPT alternatives and what each one costs.
Grok 4.7 vs GPT-6 Astra: the 3 builds
GPT-6 Astra placed above Grok 4.7 on all three builds. Grok's three runs cost about a quarter of Astra's in total, and most of that gap came from the website build, where Astra spent $97.79. Pat gave both models, plus Anthropic's Fable 5.1, the same prompt for each build, one pass each, and judged the results blind before revealing which model made which.
| Build | Who placed higher | ||
|---|---|---|---|
| Copy of an award-winning website | $7.26, 23 min | $97.79, 43 min | Astra (first of three; Grok last) |
| 3D keyboard product page | $5.73, 57 min | $5.25, 24 min | Astra (second of three; Grok last) |
| Mario Kart Rainbow Road copy | $16.54, 118 min | $11.25, 38 min | Astra (Grok last) |
| All three builds | $29.52, 198 min | $114.29, 106 min | Astra, 3 of 3 |
The costs are what each run would cost at the models' published API prices, from each build's recorded token counts. The builds themselves ran on subscriptions. Full token counts, and Fable 5.1's numbers, are in the Grok 4.7 review post.
Copying an award-winning website
Each model picked a winning site from Awwwards, a web design awards site, and rebuilt it from scratch. Pat called Astra's version the "clear winner" and Grok's "clearly the worst." Grok's run cost $7.26 against Astra's $97.79.

A 3D product page for a keyboard
Each model built a single scrolling page for a high-end mechanical keyboard, with a 3D model of the board that comes apart and gets labeled as you scroll. Fable 5.1 placed first, Astra second and Grok last ("Grok clearly the loser there," in Pat's words). Astra was also slightly cheaper here, at $5.25 against Grok's $5.73.
A copy of Mario Kart's Rainbow Road
Grok 4.7's game was the worst of the three. Pat called it "like unplayable." Astra's version had its steering reversed, so pressing left turned the kart right. Grok also worked much longer on this one (118 minutes against Astra's 38), which made it the more expensive run at $16.54 against $11.25.


Pat's read on the price gap: "having a worthy adversary show up at a fifth of the price. It's good for all of us. I just wish it was a little bit better."
Is Grok more accurate than ChatGPT?
Pat hasn't tested factual accuracy, so this page doesn't claim that either one is more accurate. His tests covered coding builds and agent jobs. SpaceXSI's model page describes Grok 4.7 as having "minimal hallucinations," which is the company's own description, and none of Pat's tests checked it.
Pat's build test also left out knowledge work, automations and everyday chat, which is where the benchmarks Grok's maker published at launch claim Grok 4.7 does well. He noted that limit in his review: three builds is a limited test.
What is Grok best used for?
In Pat's tests, Grok was best as an agent through Grok Bot, and for anything that needs X. After his three-job agent test, he said: "For Grok Bot, at least when it comes to work, it is absolutely the winner."
When his YouTube research job got to the X part, he said: "This is where Grok Bot of course will excel because of its access to X." Grok Bot can now search and watch X for you as well.
He drew one line: "The only thing I wouldn't trust it on is day-to-day coding." That fits his build test, where Grok 4.7 placed last on all three coding jobs.
Grok Bot vs ChatGPT Dots: the agent side
Grok Bot beat ChatGPT's Dots in all three of Pat's agent jobs. It found the cheaper trip ($709 for flights and $497 for an Airbnb, against Meta Muse's $801 and $633), while Dots was the slowest. In the four-hour coffee-company test, graded by Claude Opus 5.5 against an answer key, Grok Bot got an A-, Dots a B and Muse a D.
Pat on Dots: "it's not quite there yet. It feels a little half-baked, at least compared to these other two agent platforms." Dots runs on GPT-6 Astra. Grok Bot ran on Grok 4.7 in his test, a model he said "people don't love including myself." On October 6, Elon Musk said on X that Grok Bot will use "the best back end model for any given task, including Claude Opus 5.5."
Every job, grade and price is in the Grok Bot vs Dots vs Muse post.
Questions people ask
Is Grok free?
Grok's plans page lists no free plan. The cheapest is Plus at $8 a month.
Is ChatGPT free?
Yes. ChatGPT Free costs $0 and includes unlimited text chats with GPT-6, within OpenAI's abuse guardrails, with limits on uploads, images, voice and deep research.
Does ChatGPT Plus include GPT-6 Astra?
Yes. ChatGPT's plan comparison shows GPT-6 Astra on Plus ($20) and Pro (from $100), and not on Free or Go.
Which Grok plan includes Grok Bot?
Premium ($30 a month) and every plan above it, per grok.com/plans. Grok's $8 Plus plan doesn't include it.
Which ChatGPT plan includes Dots?
Only Pro, which starts at $100 a month. The pricing page lists "Dot, your always-on agent" under Pro.
Is Grok cheaper than ChatGPT?
Per token, yes: Grok 4.7's API price is $2 and $6 per million input and output tokens, against $10 and $50 for GPT-6 Astra. On subscriptions, the cheapest paid plans cost the same $8 a month, and Grok's plan with Grok Bot ($30) costs $10 more than ChatGPT Plus ($20).
Which has the better agent?
Grok Bot, in Pat's test: it got an A- overall against Dots' B, and it costs less to get, from $30 a month against $100 for ChatGPT Pro.
How Pat tested
For the builds, every model got the same prompt with nothing added, one pass, and no follow-up prompts. Each model ran inside its own lab's coding agent: Grok 4.7 in Grok Build, GPT-6 Astra in Codex and Fable 5.1 in Claude Code, all at high reasoning effort. Pat judged each build blind and then revealed the models. Token counts, cost and build time were recorded during each run.
For the agents, Pat gave Grok Bot, Dots and Meta Muse the same three jobs: plan a weekend in Montreal, research and write a YouTube video, and run a fake coffee company for about four hours. Claude Opus 5.5 graded the coffee-company job against an answer key across nine areas.
More head-to-head tests: Gemini vs ChatGPT on the same 4 builds.
Update log
November 5, 2026: Page published. ChatGPT and Grok plan prices checked on chatgpt.com/pricing and grok.com/plans; API prices checked on SpaceXSI's and OpenAI's pricing pages.






Comments
Sign in to join the conversation.