Sort new products into the right category in your shop

Embeddings shortlist the most plausible categories for each new product; a small model picks one and explains why — workable even on large category trees.

How it works: Your category tree is embedded once with BGE-M3. For each new product, embeddings and the reranker shortlist the five to ten most plausible categories, and the small LLM picks one and explains why. Shortlisting first keeps the task small enough for the fast model to handle even on trees with hundreds of nodes — it never has to consider your whole taxonomy at once.
Data you need: Your full category tree with names, plus product titles and descriptions. One-line descriptions per category are optional but noticeably improve the shortlist — most shops do not have these written down, and writing one line each is a small one-off job worth doing first. A few hundred already-correctly-categorised products help as reference examples but are not strictly required.
What to expect: Reliable on products whose category is clear from the text, and it consistently explains its choices, which makes review fast. If two categories overlap heavily, the model will be inconsistent between them exactly as your staff are — it will not fix a badly designed taxonomy, it will faithfully reproduce its problems. Bare category names like 'Accessories' or 'Other' give it little to go on. It works on text only: if a product's category is only obvious from its photos, the model is guessing from the title and description.
Where people stay involved: Genuinely ambiguous products (is a yoga towel 'Yoga' or 'Towels'?) reflect ambiguity in your tree, not model failure — route low-confidence picks to a person, and review a random sample of the confident ones periodically.

Which model, and what it costs to run

Qwen3-32B

This job runs in bulk rather than to a waiting person, so size is not the constraint — we take the strongest independent score on structured output and tool calls that we may serve freely and that fits on a single card.

Licence: Apache-2.0
Weights at 4-bit: 18 GB
Context: 32K tokens
Publisher: Alibaba (Qwen Team)

What the hardware costs

One 48 GB card holds it

Rent in the EU: $1.60/hrScaleway, Paris (PAR2)
Buy the card: $7,569new, one-off
Or rent it by the token: $0.12 / $0.12per M in / out · Nebius AI Studio · EU

Hardware only, third-party prices from 2026-07. The figure excludes the KV cache, which grows with context length and how many people use it at once — sized properly in a conversation, not guessed here. Renting by the token is cheaper up front; why our customers still self-host is below.

Structured output and tool calls

Berkeley Function Calling Leaderboard · v4 · 13 of 18 models measured

Measured in a sandbox, on somebody else's functions. It tells you which models are capable of the shape of the job, not which one will survive contact with your API.

Longer is betterFree to serveConditions apply⚠ answered under 95%

Independent measurement · UC Berkeley (Gorilla project) · board updated 2026-04-12

Writing in your voice

No benchmark measures this

There is no benchmark for this and there is unlikely ever to be one. Anyone who shows you a chart ranking models on marketing copy has drawn it from a model's opinion of another model.

So we do not show a chart here. We measure it on your own content, in the first week, and you see the result before anything ships.

The API is cheaper per token. Here is why our customers don't use it.

We will not pretend otherwise: renting a model by the token from a serverless API costs less per million tokens than a card we run for you. We show that price on every use-case page. What it does not include is the part a shop with a customer database actually pays for.

01
Your data never leaves hardware you can point at
A serverless "we don't retain your data" is a clause in a contract. Running the model on a card in Amsterdam is a fact of architecture: your catalogue, tickets and customer records are never sent to a third party at all. For a GDPR audit, that is the difference between a promise and a floor plan.
02
The price cannot move without your say-so
A serverless rate card is somebody else's lever. The provider can raise the price, retire the model, or change the terms, and your cost moves with it. The same model on the same card costs the same next year — you own the number.
03
The model cannot be taken away
Hosted APIs deprecate models on their own schedule; the one you built on can be gone in a quarter. An open-weight model on your own hardware runs for as long as you keep the lights on. No vendor can end-of-life it out from under you.

And the price gap closes with volume: past a card you keep busy — very roughly four billion tokens a month — owning is cheaper outright, even before the three reasons above.

Getting a case like this one from a conversation to production takes about two months, and you can stop at the end of any phase.

Also in catalogue content

All 14 →

Sort new products into the right category in your shop

Qwen3-32B

Structured output and tool calls

Writing in your voice

The API is cheaper per token. Here is why our customers don't use it.

Also in catalogue content

Turn supplier spec sheets into first-draft product descriptions

Pull sizes, materials, colours and other attributes out of messy supplier text

Write page titles and meta descriptions for every product and category page

Draft alt text for product images from the product data you already have

Tell us what your team does by hand