Grok 4.7 vs ChatGPT GPT-6 Astra: Ten Prompts Tested Blind
GPT-6 Astra received nine of ten selections in this matched prompt set; Grok 4.7 received one.
In ten matched prompts, GPT-6 Astra received nine selections and Grok 4.7 received one. Malte Prüser was the only blind reviewer. This is a result for one prompt set, not a general ranking of Grok and ChatGPT.
Results across ten prompts
| Prompt | Recorded selection | Observable difference in the outputs |
|---|---|---|
| 1. Customer email | GPT-6 Astra | GPT marked Friday as expected, not guaranteed; Grok added carrier and follow-up details. |
| 2. Product update | GPT-6 Astra | Both retained the retry process and rollout conditions; their explanations differed in length. |
| 3. Python function | GPT-6 Astra | Both supplied a top-scorer function and tests; GPT included negative-score coverage. |
| 4. Bug review | GPT-6 Astra | Grok filters nonnumeric values and returns None when none remain; GPT raises on empty input. |
| 5. Google AI Pro | GPT-6 Astra | Grok returned an intent to check; GPT gave remembered details with a live-check caveat. |
| 6. ChatGPT Plus trial | GPT-6 Astra | Grok returned an intent to check; GPT described limited offers and acknowledged no live check. |
| 7. Statistics | GPT-6 Astra | Both gave mean 20, median 14, and range 28; GPT included the counterfactual mean. |
| 8. Team seat cost | Grok 4.7 | Both calculated $395 and $39.50 per person and flagged the inputs as hypothetical. |
| 9. Bakery image prompt | GPT-6 Astra | GPT specified a three-meter camera distance and 35mm lens; both gave scene direction. |
| 10. Water-cycle prompt | GPT-6 Astra | GPT named stages including transpiration and infiltration; both requested no readable text. |
Quick read
The nine-to-one count summarizes this blind ranking, but it should not be stretched into a claim about every task or feature. Grok's single selection came on arithmetic using a fixed set of supplied prices. GPT-6 Astra received the selections on the email, product update, code, bug review, two source prompts, statistics, bakery prompt, and water-cycle prompt.
The ranking is not a subscription comparison; what happened to SuperGrok pricing examines the current Super price, annual payment, and legacy-plan caveat.
When you read the task notes, separate three things: the blind selection, what the response visibly contains, and whether any factual statement was checked outside the test. A model can sound confident without providing an official source. For that reason, current prices and trial terms below come from vendor pages checked for this article, not from Arena answers.
The test did not compare Grok and ChatGPT subscriptions feature by feature. It also did not send one prompt to every available model, test long documents, or inspect an image generated from the two written prompts. The price table is a separate purchasing reference.
How we tested
We ran the unchanged ten-prompt set in Manual Model Arena on November 8, 2026. Both model entries received each prompt with default temperature and no system prompt. The responses were anonymized and shuffled as A and B. Malte Prüser was the only blind reviewer. The ranking file contains one selection per pair and no written rationale, so the observations below describe the saved text rather than reasons assigned to the reviewer. See our model testing methodology for the broader procedure.
| Picker name | Menu ID | Version note |
|---|---|---|
| Grok 4.7 | x-ai/grok-4.7 | 4.7 as displayed in the picker |
| GPT-6 Astra | openai/gpt-6-astra | Exact picker name; no further upstream version exposed |
The Arena is one tool in Krater, The All-in-One AI Workspace with 500+ AI models. This run used the two exact menu entries above, not a subscription checkout or every model in those companies' product catalogs. The recorded response time and credits are per paired battle. No number below is attributed to Grok or GPT individually.
| Prompt | Paired response time | Paired credits |
|---|---|---|
| 1. Customer email | 14.1921 seconds | 7 |
| 2. Product update | 68.9271 seconds | 15 |
| 3. Python function | 9.7689 seconds | 5 |
| 4. Bug review | 11.5294 seconds | 4 |
| 5. Google AI Pro | 15.6459 seconds | 7 |
| 6. Trial policy | 20.0156 seconds | 8 |
| 7. Statistics | 8.5820 seconds | 3 |
| 8. Seat cost | 6.1849 seconds | 2 |
| 9. Bakery prompt | 10.9877 seconds | 3 |
| 10. Water-cycle prompt | 12.6391 seconds | 3 |
Task 1: Customer email
Prompt: Draft an email about order 8421, delayed by a carrier disruption for two days, with Friday as the expected date and a refund available on request. Do not invent promises.
Grok adds that the carrier provided no more details, that no time of day can be guaranteed, and that a further update will be sent if the date changes. Those details were not in the prompt. GPT-6 Astra also distinguishes an expected date from a guarantee, and makes the refund option explicit without describing an order-processing policy. The recorded selection was GPT-6 Astra. The examples are drafts, not facts about a real shipment.
Task 2: Product update
Prompt: Explain two retries after a failed upload, storing an error code, an unchanged user data model, a 20 percent rollout, and stopping if failure rates rise.
Grok defines the sync worker in simple terms and explains that the error code records the failure. GPT explains the background sync process and restates that the data model is unchanged. Both preserve the two retries and rollout condition. The answers vary in the amount of repetition and in how they explain the technical term. The saved selection was GPT-6 Astra. No release date or extra capability is inferred from the rewrite.
Task 3: Python function and tests
Prompt: Return names with the highest score in alphabetical order, handle an empty list, include three tests, and explain complexity.
Both responses compute the maximum, collect records with that value, and sort the resulting names. Grok's sample tests include an empty list, a single highest scorer, and a tie. GPT's examples also cover a tie and an empty list, with a negative-score case. Both describe a scan plus sorting the tied names. The recorded selection was GPT-6 Astra. We reviewed the code statically and did not execute these generated snippets.
Task 4: Bug review
Prompt: Fix the empty-input bug in average(values) without changing the signature, and state how it handles empty and nonnumeric values.
Grok builds a new list containing only integers and floats, excludes booleans, and returns None if no valid numbers remain. That silently discards invalid items. GPT raises ValueError on an empty collection and allows sum to raise TypeError for ordinary nonnumeric values. These are different policies, not interchangeable definitions. The underlying expression does divide by zero on an empty list. The recorded selection was GPT-6 Astra.
Task 5: Google AI Pro price and sharing
Prompt: Use official sources to report the current U.S. price and whether Google AI Pro's plan page mentions family sharing.
Grok's saved answer says it will check official pages but supplies no price or conclusion. GPT-6 Astra gives $19.99 per month and a family-sharing statement, while warning that it cannot confirm live wording in that chat. We checked the current Google pages separately: the U.S. Google One page lists the $19.99 price and 5 TB; the Google AI plan page says family sharing is available with up to five others. The answer excerpt is not the citation. GPT-6 Astra received the recorded selection.
Task 6: ChatGPT Plus trial policy
Prompt: Determine whether OpenAI describes a universal Plus free trial or limited eligible promotions, with exact official sources.
Grok again returns an intent to check without providing the requested policy answer. GPT-6 Astra says OpenAI describes limited, eligibility-based offers and not an entitlement for every account, but notes it cannot live-check. OpenAI's current Help Center independently describes promotional subscriptions and trial invitations for eligible users, with campaign-specific terms. The 2026 Back to School offer is separate and its October 31 claim deadline had passed at the time of this article. GPT-6 Astra received the recorded selection.
Task 7: Mean, median, and range
Prompt: Calculate statistics for 12, 14, 14, 20, and 40 and explain sensitivity to the largest value.
Both answers report a mean of 20, median of 14, and range of 28. GPT also shows that removing 40 lowers the mean to 15. Grok notes that changing the largest value changes the mean while the median can remain the same. These calculations are verified directly from the inputs. The recorded selection was GPT-6 Astra.
Task 8: Team seat cost
Prompt: Calculate eight $20 seats, two $100 seats, and a $35 platform fee. Divide by ten people and say the numbers are supplied inputs, not a quote.
Both answers give $395 total and $39.50 per person. Grok places the totals first and states the required caveat immediately. GPT shows the same component arithmetic and caveat. This is a hypothetical arithmetic problem; it is not a statement of either company's current plan pricing. The recorded selection was Grok 4.7.
Task 9: Bakery image prompt
Prompt: Write a photorealistic editorial prompt for a small bakery preparing for the morning rush, specifying composition, light, and camera distance.
Grok describes a bakery seen from the entrance and mixes warm interior light with blue dawn light. GPT gives a customer-side view from about three meters away, a 35mm lens, and a counter leading toward the work area. Both avoid readable brands and describe ordinary bakery details. The Arena judged the written prompt, not the image it might produce. GPT-6 Astra received the recorded selection.
Task 10: Water-cycle diagram prompt
Prompt: Create a clean middle-school water-cycle diagram prompt with a clear layout and no readable text.
Grok maps ocean, land, river, mountains, clouds, rain, runoff, and groundwater. GPT includes water rising from a lake, clouds and rain, a river, groundwater, and a tree with an upward arrow for transpiration. USGS describes transpiration and infiltration as parts of water movement between land, plants, and atmosphere. This checks terminology only; no image was generated. GPT-6 Astra received the recorded selection.
Current Grok and ChatGPT plan prices
The subscriptions below were checked on November 8, 2026. They explain consumer access and are not the model configuration used by the Arena. The X product has moved to a new self-serve bundle, so older pages that still say SuperGrok may describe a legacy plan. xAI says existing SuperGrok, Cursor, and X Premium subscribers can continue unless they choose to switch.
| Product | Current individual plan prices | Notes from the official page |
|---|---|---|
| X Subscription | Plus $8/mo or $80/yr; Premium $30/mo or $300/yr; Super $100/mo or $1,000/yr; Ultra $200/mo or $2,000/yr | Usage is pooled across supported Grok and Cursor features; third-party models in Cursor are billed separately. |
| ChatGPT | Go $8/mo; Plus $20/mo; Pro $100, $200, or $500/mo | Prices shown are U.S. monthly prices; check the account checkout for local availability and taxes. |
xAI's annual page shows rounded monthly equivalents of $7, $25, $84, and $167 for those annual plans. Its current pricing page says that included AI usage refreshes weekly and varies by product, so a subscription price does not equal a fixed number of prompts. OpenAI lists three monthly Pro tiers; the official plan details describe different allowances rather than one flat Pro limit. Sources: xAI pricing, X Subscription prices, ChatGPT pricing, and ChatGPT Pro tier details, checked November 8, 2026.
Frequently Asked Questions
How many prompts did GPT-6 Astra win?
GPT-6 Astra received nine of ten recorded selections, and Grok 4.7 received one. The table shows the prompt-by-prompt result.
Was Grok 4.7 selected on a task?
Yes. Grok 4.7 received the selection on the supplied-input team seat-cost calculation. Both responses gave the same $395 total and $39.50 average.
Was this a two-reviewer test?
No. Malte Prüser was the only blind reviewer. The ranking file has one selection for each of ten prompt pairs and no second set of votes.
Does the result compare current Grok and ChatGPT subscriptions?
No. It compares picker entries. The subscription table above is current plan information checked separately and does not explain an individual task result.
Did the test measure speed per model?
No. The 178.47 seconds and 57 credits are totals across paired battles. The run data does not provide a separate response-time or credit total for either model.
Related reads
Bottom line
GPT-6 Astra received nine selections and Grok 4.7 received one on this ten-prompt set. The raw responses include useful differences in how each handles unsupported details, code, explicit caveats, and formatted prompts. Teams should use the examples to design their own evaluation tasks, while checking current prices and policy claims against the vendor pages. One round cannot establish a universal winner.