GPT Image 2.5 FlareFrom $0.050/imageTwo real API calls show how a text-only reasoning model handles catalog copywriting and policy-aware customer support at sub-cent cost per request.
Choose a model, enter your prompt, and see the result.
HiAPI Blog
HiAPI
Generate it with HiAPI
E-commerce teams generate a constant stream of short, structured text: product titles, bullet points, size-guide answers, return-policy replies, review responses. None of it needs a general-purpose chat model burning tokens on multi-page reasoning — but it does need to be fast, consistent, and cheap enough to run on every SKU and every support ticket.
kimi-k3, available through the hiapi API, is a text-only reasoning model built for exactly this kind of workload. It has no image or video output — if you came here looking for AI-generated product photography, see the gpt-image-2 e-commerce workflow guide instead. What kimi-k3 is genuinely good at is turning a messy product spec or a testy customer message into clean, on-brand text, with a selectable reasoning budget so you only pay for the thinking you need.
This guide covers two concrete e-commerce use cases — product copywriting and customer support replies — with real API calls, real token usage, and real per-request cost.
Most e-commerce text tasks are single-turn and well-scoped: rewrite this spec sheet into five bullet points, answer this one return question, draft a product title within a character limit. You don't need vision, you don't need multi-step tool use — you need a model that follows formatting instructions reliably and doesn't hallucinate policy details.
kimi-k3 exposes three reasoning_effort levels (low, high, max) on the same Chat Completions endpoint, so you can dial reasoning up for messages that require actually applying a policy (a return eligibility check) and dial it down for tasks that are mostly formatting (turning spec bullets into marketing copy). That's a cost lever most single-tier chat models don't give you.
A typical catalog job: take a raw spec dump — dimensions, materials, a couple of feature notes — and turn it into a title plus bullet points that fit a marketplace character limit and don't overstate what the product does.
curl https://api.hiapi.ai/v1/chat/completions \
-H "Authorization: Bearer $HIAPI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "kimi-k3",
"reasoning_effort": "low",
"messages": [
{
"role": "system",
"content": "You write e-commerce product listings. Output a title under 70 characters and 4 bullet points in ALL-CAPS-LEAD style. Do not claim features not in the spec."
},
{
"role": "user",
"content": "Product: pour-over coffee dripper. Material: stainless steel. Capacity: 1-2 cups. Color: matte black. Includes travel pouch. Dishwasher safe."
}
]
}'
With reasoning_effort set to low, the response comes back in well under a second of model "thinking" and produces a ready-to-paste listing:
Product Title (69 characters): Stainless Steel Pour Over Coffee Dripper, 1-2 Cup Cone Filter, Matte Black
- PRECISION POUR-OVER BREWING — Cone-shaped stainless steel dripper designed for 1–2 cups, giving you full control over extraction for a clean, flavorful cup every time.
- DURABLE STAINLESS STEEL — Built from stainless steel with a sleek matte black finish that looks great on any kitchen counter or coffee bar.
- REUSABLE CONE FILTER DESIGN — No paper filters needed for the dripper itself; enjoy a sustainable, cost-saving brew routine at home or on the go.
- EASY CLEAN & TRAVEL READY — Dishwasher safe for effortless cleanup, and the included travel pouch makes it simple to brew your favorite coffee anywhere.
Note the model's own caveat appended after the bullets, flagging that the "reusable filter" claim depends on a spec detail it wasn't given (whether the dripper needs paper filters). That kind of self-flagged uncertainty is exactly what you want in an automated copy pipeline — it's a signal to route the item to human review rather than publish an unverified claim.
Real usage from this call: 184 prompt tokens, 239 completion tokens (including 10 reasoning tokens), for a total cost of $0.0032.
Support questions need more actual reasoning than copywriting — the model has to apply a policy correctly, not just reformat text. This is where a higher reasoning_effort earns its keep.
curl https://api.hiapi.ai/v1/chat/completions \
-H "Authorization: Bearer $HIAPI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "kimi-k3",
"reasoning_effort": "high",
"messages": [
{
"role": "system",
"content": "You are a support agent. Return policy: returns accepted within 30 days if unused and in original packaging. Customer pays return shipping unless the item arrived damaged or wrong. Refunds post 5-7 business days after warehouse receives the item."
},
{
"role": "user",
"content": "I ordered the wrong size 10 days ago, item is unopened. Can I return it, and who pays shipping?"
}
]
}'
Response:
Good news — you're eligible for a return! Here's how it works based on our policy:
Return eligibility: Yes, you can return it. Returns are accepted within 30 days as long as the item is unused and in its original packaging. Since you ordered it 10 days ago and haven't opened it, you qualify.
Return shipping: Since the item arrived correctly (the wrong size was ordered rather than sent by mistake), return shipping would be your responsibility. We only cover return shipping costs when an item arrives damaged or we sent the wrong item.
Refund timing: Once our warehouse receives your return, your refund will post to your original payment method within 5–7 business days.
The model correctly distinguishes "customer ordered the wrong size" from "we shipped the wrong item" — a distinction that changes who pays for return shipping — and doesn't just pattern-match on the word "return." That's the kind of policy application you'd want checked before this goes out as an auto-reply, but it's a solid first draft that a human agent can approve in seconds instead of typing from scratch.
Real usage from this call: 222 prompt tokens, 207 completion tokens (including 45 reasoning tokens — more than double the copywriting call, reflecting the extra policy-reasoning work), for a total cost of $0.0029.
kimi-k3 pricing on hiapi is $2.32 per million input tokens and $11.60 per million output tokens, with cached input billed at $0.232 per million — a 90% discount versus fresh input tokens. If your support system reuses the same policy text as a system prompt across thousands of tickets, that system prompt becomes a strong candidate for prompt caching: only the customer's message and the model's reply cost full price, and the repeated policy block gets billed at cache rates on cache hits.
For copywriting at catalog scale, reasoning_effort: "low" keeps completion tokens (and therefore cost) down for tasks that are mostly reformatting rather than reasoning. Reserve high or max for tasks where getting the logic wrong (a refund calculation, a warranty exception) is more expensive than a slightly slower response.
Full parameter reference, including response_format for structured JSON output and streaming support, is in the kimi-k3 docs.
Both examples above run against the same kimi-k3 model page and endpoint — swap the system prompt and reasoning_effort value to match your task, and you have a working copywriting or support-reply pipeline in one API call. Check current pricing for all available models, grab an API key, and start with a handful of real product specs or real (anonymized) support tickets rather than synthetic test data — the model's output quality is easiest to judge against the messy, real-world text it'll actually see in production.
