TL;DR
- Five ecommerce and office tasks, each with GPT-6.1 Terra's real output, the full prompt and what the run cost.
- Each run cost between $0.0001 and $0.001, so $1 covers hundreds of runs of every task here.
- Copy a prompt and paste it into the GPT-6.1 Terra model page to run it. New accounts get up to 2,000 free credits ($1).
Task 1: write a product listing from a photo
{"product_name":"Orange Robot Plush Toy","category":"Plush toys","material":null,"main_color":"Orange","selling_points":["Friendly robot face with large eyes","Orange antenna and rounded limbs","Lightning bolt emblem on the front"],"target_buyer":null}
This is a product photo. Fill in the listing fields in the given JSON schema: product name, category, material, main color, 3 selling points, target buyer. Use null for anything you cannot see in the photo; don't guess.
Copy the prompt, open the model page and upload a photo →
gpt-6.1-terra, medium, image + strict JSON schema, 436 input / 61 output tokens, 6 seconds, $0.0002.
The line "use null for anything you cannot see" does the work here. Material and target buyer can't be read from a photo, so the model left them empty instead of inventing them. The schema setup is in the API guide.
Task 2: turn meeting notes into action items
| Task | Owner | Due date | Status |
|---|---|---|---|
| Create Black Friday landing page design draft | Lily | Nov 5 | Pending |
| Investigate the rise in payment failure rate and report the cause | Tom | Wednesday | Pending |
| Move the refund button to the order detail page | TBD | TBD | Pending |
| Test app v3.2 before it goes to review | Grace | Before next Thursday | Pending |
| Add a repeat purchase rate metric to the dashboard | Alex | This week | Pending |
Turn these meeting notes into an action item table with columns: task, owner, due date, status. Mark missing owners or dates as "TBD".
Notes: Monday product sync. 1) Black Friday landing page: Lily owns it, design draft due Nov 5. 2) Payment failure rate rose to 2.3% last week; Tom will find the cause and report by Wednesday. 3) Support says the refund button is buried too deep and should move to the order detail page; no owner yet. 4) App v3.2 goes to review next Thursday; Grace owns testing. 5) The dashboard needs a repeat purchase rate metric; Alex says it'll be done this week.
Copy the prompt and try it on the model page →
gpt-6.1-terra, medium, 151 input / 116 output tokens, 5 seconds, $0.0002.
With the columns and the missing-value rule spelled out, the table pastes straight into Notion or a spreadsheet. Item 3 had no owner, and the model marked it TBD as asked.
Task 3: reply to an upset customer
I'm sorry you've waited 12 days, especially with no tracking update for five. We do care, and I understand how frustrating this is.
We can either reship your order or issue a full refund—whichever you prefer. Please reply with your order number and choice, and we'll follow up within one business day to confirm the next steps.
You are a support agent for an online store. Reply to this customer: apologize, offer a resolution (reship or refund, their choice), give a timeline, keep it sincere and under 100 words.
Customer: I've waited 12 days and my package still hasn't arrived. Tracking hasn't updated in 5 days. Does anyone there care? If it doesn't show up soon I'm filing a complaint.
Copy the prompt and try it on the model page →
gpt-6.1-terra, low, Chat Completions with streaming, 86 input / 76 output tokens, first token at 2.5 seconds, done at 4.5, $0.0001.
Low effort is enough for short replies. The "one business day" timeline came from the model, not from you, so put your store's real policy in the prompt before using this with customers.
Task 4: extract an order from a customer email
{"order_id":"HX-20481","order_date":"Oct 20","items":[{"name":"hoodie","color":"dark grey","size":"L","qty":2},{"name":"joggers","color":"black","size":"M","qty":1}],"total":"$86.40","issue":"Both hoodies arrived light grey instead of dark grey; the black joggers are correct.","requested_action":"Exchange both hoodies for dark grey ones, or provide a refund if dark grey hoodies are out of stock.","phone":"+1 415 555 0123","contact_time":"Afternoons"}
Extract the order details from this customer email.
Email: Hi, I placed an order last Tuesday (Oct 20), order number HX-20481: two dark grey hoodies in size L and one pair of black joggers in size M, $86.40 in total. The package arrived today and the hoodies are light grey; the joggers are fine. I'd like to exchange both hoodies for dark grey, or get a refund if you're out of stock. You can reach me at +1 415 555 0123, afternoons work best. - Sarah
Copy the prompt and try it on the model page →
gpt-6.1-terra, medium, strict JSON schema, 236 input / 185 output tokens, 23 seconds, $0.0003.
Order number, total, phone and contact time all came through exactly, and the color field holds the ordered color while the mismatch went into issue. The schema set the fields; the prompt itself is one line.
Task 5: write Amazon listing copy
Title: Portable Blender, 380 ml/13 oz, USB-C Rechargeable, 6-Blade, IPX7 Waterproof
- Blend smoothies, shakes, and drinks anywhere with a compact 380 ml Tritan cup and lightweight 420 g design.
- USB-C charging delivers up to 15 blends per charge for convenient use at home, work, or on the go.
- Six stainless steel blades support smooth blending, while IPX7 waterproofing lets you rinse the whole unit after use.
Write Amazon product listing copy from these specs: a title under 80 characters and 3 bullet points, each under 25 words. Avoid words like "ultimate" or "premium".
Specs: portable blender, 380 ml (13 oz), USB-C charging, 15 blends per charge, 6 stainless steel blades, Tritan cup, 420 g, IPX7 waterproof so the whole unit can be rinsed.
Copy the prompt and try it on the model page →
gpt-6.1-terra, medium, 92 input / 693 output tokens (585 of them reasoning), 20 seconds, $0.0009.
The title is 76 characters, each bullet is 18–19 words, and neither banned word shows up. Most of the cost was reasoning, not the copy itself: for batches of listings, try low or none to cut both cost and wait time.
How to write prompts for Terra
- Fix the format. Name the columns, word limits or JSON fields, and the output is ready to use.
- Say what to do when it doesn't know. "Use null" or "mark as TBD" is more reliable than leaving it open.
- Supply the business facts. Put timelines, prices and policies in the prompt so the model doesn't fill them in.
FAQ
Which reasoning effort should I use for these tasks?
Low or medium for extraction, cleanup and short replies; try none for short copy. Costs by effort are in GPT-6.1 Terra API pricing.
Can I upload images on the model page?
Yes. The Playground takes up to 4 images per message, 10 MB each, sent along with your text.
How do I run these in bulk?
Use the API with the same key: POST /v1/responses, with code in the API guide. To compare results against Sol, see GPT-6.1 Terra vs GPT-6.1 Sol.
Pick the task closest to your own work, copy its prompt, and open the GPT-6.1 Terra model page to run it with your own data.
