claude-opus-4-8 shows up on hiapi tagged LLM, Reasoning, Vision, Streaming, Tool Calls. No Image Generation tag anywhere — it does not produce pictures. So when a brief calls for "product images," the honest answer is: wrong model. What it's actually good for in an online store is reading images and writing text: grading product photos against your own QA rules, writing product copy, and answering support questions. Below are three real calls against the hiapi API, with the real output and the real cost for each.
Why claude-opus-4-8 for e-commerce text and vision work
claude-opus-4-8 is available on hiapi through two routes that point at the same model with different pricing:
claude-opus-4-8(default) — the cheaper route.claude-opus-4-8-aws— routes through AWS infrastructure, and costs exactly 2.5× the default route on both input and output tokens.
For e-commerce copy, support replies, and photo QA, none of which need anything the AWS route offers beyond routing, the default route is the one to reach for. Full current numbers are on the hiapi pricing page — check it before you commit to a route, since per-token pricing is exactly the kind of number that moves over time.
One thing worth flagging before the examples: the default route's own model card describes it as text generation, streaming, and tool calling — it doesn't mention image input at all. The claude-opus-4-8-aws card does, explicitly: "supports text and image input." In the test below, the default route handled an image input request correctly anyway. The lesson isn't "trust the vision tag over the card" or "trust the card over the tag" — it's that when a model card and a capability tag disagree, the only reliable answer is to send a real request and read the real response, which is what the rest of this article does.
Use case 1: Product photo QA via Vision
The idea: instead of a human scanning every new listing photo, hand the model a short rule set and a photo, and let it flag anything that fails before the listing goes live.
The test photo here was a genuinely bad case on purpose: a flat-lay of about ten unrelated accessories (wallet, sunglasses, bow tie, pen, notebook, phone, watch, headphones) on a dark background, not a clean single-product shot. The goal was to see whether the model would actually catch that, rather than rubber-stamp it.
curl https://api.hiapi.ai/v1/chat/completions \
-H "Authorization: Bearer $HIAPI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-opus-4-8",
"messages": [
{
"role": "system",
"content": "You are a marketplace photo QA reviewer. Check the product photo against these four rules and give a PASS or FLAG verdict for each, with a one-sentence reason: (1) product is centered and fully visible, (2) background is clean and neutral, (3) no visible damage or defects, (4) lighting is even enough to judge true color. End with one overall verdict."
},
{
"role": "user",
"content": [
{"type": "text", "text": "Review this listing photo."},
{"type": "image_url", "image_url": {"url": "https://picsum.photos/id/26/800/600"}}
]
}
]
}'
Actual response, unedited:
Rule 1 — Centered and fully visible: FLAG. This isn't a single-product shot. It's a flat-lay of roughly ten items (zip wallet, sunglasses, bow tie, pen, card case, notebook, card holder, phone, watch, glasses, headphones) spread across a grid with wide empty margins. No item is centered, and any listing using this would have no clear subject. If one product is the actual listing, it needs its own crop.
Rule 2 — Clean/neutral background: PASS. Seamless dark gray surface, uniform across the frame, no competing props or patterns. Same tone holds into all four corners.
Rule 3 — No visible damage or defects: PASS (limited). Nothing reads as cracked, scuffed, or torn, and no packaging appears in frame at all. Worth noting the assessment is limited by resolution and scale — small scratches or stitching flaws on the leather goods wouldn't be detectable here, so if damage checking matters for this SKU, use a tighter shot.
Rule 4 — Lighting shows true color: FLAG. The exposure is dim and the shadow falloff is heavy, especially in the lower-left and along the bottom edge. The blacks don't resolve as black — the wallet, card holder, and phone read slightly warm/olive, and the "green" edge on the notebook is ambiguous. A buyer can't reliably judge color from this. Brighter, more even, higher-key lighting would fix it.
Overall: FLAG — do not publish as-is. The frame is a multi-item lifestyle composition, not a compliant single-product photo, and it's too underexposed for accurate color. Re-shoot as one centered product on the same neutral background with even lighting, or crop a single item and re-expose.
It caught the actual problem (multiple products, no clear subject) instead of just describing the picture, and it hedged correctly on rule 3 instead of overclaiming certainty from a low-resolution image — a caveat most rubric-checking scripts wouldn't think to add on their own.
This call used 769 tokens (143 prompt + 626 completion), which comes out to about $0.0075 on the default route.
Use case 2: Product copywriting
Same model, plain text this time — turn a spec sheet into a listing.
curl https://api.hiapi.ai/v1/chat/completions \
-H "Authorization: Bearer $HIAPI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-opus-4-8",
"messages": [
{
"role": "system",
"content": "You are an e-commerce copywriter. Given a product'"'"'s specs, write: a title (max 70 characters), 3 bullet feature points, and a 2-3 sentence description. No superlatives, no exclamation marks, stay factual."
},
{
"role": "user",
"content": "Product: zip-around wallet. Material: full-grain leather, hand-stitched edges. Capacity: 6 card slots + cash compartment. Closure: zip-around. Fit: slim enough for a front pocket."
}
]
}'
Actual response, unedited:
Title: Slim Zip-Around Full-Grain Leather Wallet — 6 Cards + Cash
Features:
- Full-grain leather with hand-stitched edges that hold up to daily use
- Holds 6 cards plus folded cash in a low-profile frame
- Zip-around closure keeps contents secure; sized to fit a front pocket
Description: This zip-around wallet carries six cards and cash in a build slim enough for your front pocket. Full-grain leather and hand-stitched edges mean it ages with wear and stays intact through years of everyday carry.
No "premium," no "must-have," no exclamation marks — it followed the "no superlatives" instruction, which is usually the part copywriting prompts fail at. 364 tokens total, about $0.0026.
Use case 3: Customer support replies
Feed it your actual return policy and a real customer message, and it drafts a reply a human can send with light editing (or none).
curl https://api.hiapi.ai/v1/chat/completions \
-H "Authorization: Bearer $HIAPI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-opus-4-8",
"messages": [
{
"role": "system",
"content": "You are a customer support agent for an online wallet store. Store policy: 30-day return window from delivery; if the item arrived damaged or wrong, the store covers return shipping and offers a replacement or exchange. Reply in under 150 words, be empathetic but concise, and end with a clear next step."
},
{
"role": "user",
"content": "I ordered a wallet 3 weeks ago and it arrived with a scratch on the front panel. Can I still return it, and do I have to pay for shipping?"
}
]
}'
Actual response, unedited:
Sorry about that — a scratch on a brand-new wallet is a letdown, and you're right to expect better from us.
Both answers work in your favor:
- Return window: you're 3 weeks in, so you're inside our 30-day policy.
- Shipping: a scratch on arrival counts as damage, so we cover return shipping. You pay nothing to send it back.
What to do: send your order number and a photo of the scratched front panel. We'll issue a prepaid label and get the replacement moving.
If you'd rather switch to a different wallet instead of a straight replacement, say the word and we'll handle the price difference.
It applied both policy rules correctly (window + damage-shipping) instead of just picking one, and it stayed under the 150-word cap. 385 tokens, about $0.0028.
Cost notes
All three calls above ran on the default claude-opus-4-8 route and totaled 1,518 tokens for roughly $0.013 combined — no image generation, no per-image charge, just token cost on the model's own pricing page. Vision input on this model is billed as tokens like any other input, not a flat per-image fee, so a bigger or busier photo (more items, more detail to describe) will cost more than the flat-lay test above. If you're running this at listing-catalog scale, budget per token, not per image, and check current rates before you commit to a monthly volume — they're exactly the kind of number that moves.
FAQ
Can claude-opus-4-8 generate product images? No. It's tagged Reasoning and Vision, not Image Generation, and every response it returns is text. If the actual job is producing new product photography or lifestyle shots, that needs an image model, not this one.
Does claude-opus-4-8 actually support image input on hiapi?
Yes — confirmed by sending a real image in the request above and getting back an image-grounded response, on the default route. The claude-opus-4-8-aws route's own model card documents image input explicitly; the default route's card doesn't mention it, which is worth knowing before you assume a card is the full story.
Which route should I use, default or AWS?
For text and vision e-commerce workloads like the three above, the default claude-opus-4-8 route — it's the same model at 1/2.5 the per-token price of claude-opus-4-8-aws.
How much does it cost to QA one product photo? About $0.0075 for the flat-lay test above (769 tokens). A simpler single-product photo with a shorter rule set would cost less; a denser image or a longer rubric would cost more, since vision input is billed as tokens.
Getting started
To try this yourself: an hiapi API key, the claude-opus-4-8 model page for the default route's current specs, or claude-opus-4-8-aws if you specifically need that routing. If you're building out a broader e-commerce content workflow on hiapi, the kimi-k3 e-commerce writeup covers the same copywriting-and-support ground with a different model, worth comparing against the results above before you pick one for production.
