ML Env

Use casesCustomer service

Sort the support inbox before anyone reads it

Classifies every incoming message into your own categories, pulls out order details, and routes it to the right queue before anyone opens it.

How it works
Every incoming email or contact-form message is classified into categories you define — where-is-my-order, return request, damaged item, invoice question, wholesale enquiry, spam — and routed to the right queue or person. The model also pulls out the order number, customer email and product mentioned, so the ticket arrives pre-filled. Sorting text into a fixed set of categories is the kind of task a small, fast model is well suited to, which keeps it cheap enough to run on every message.
Data you need
A live feed of incoming messages (IMAP, helpdesk API or a webhook) and a written list of your categories with two or three example messages each. Every shop already has the messages; the only new work is writing the category list down.
What to expect
Routine messages land in the right queue with usable pre-filled fields, and correcting the occasional wrong label is cheap. Ambiguous messages — a complaint that is also a return request — get one label when they deserve two, unless you allow multiple labels. Categories that overlap in meaning will be confused with each other; the fix is usually rewriting the category definitions, not a bigger model. The model reads text only: photo attachments, typical for damaged-item claims, pass through unread, so those tickets are sorted purely on what the customer wrote.
Where people stay involved
Humans still answer everything; the model only sorts and pre-fills. Review the routing weekly at first — a miscategorised ticket is annoying but cheap to correct. Treat extracted fields such as the order number and product as suggestions to confirm, not facts: the model can occasionally misread or invent one when the message is vague.

Which model, and what it costs to run

Qwen3-32B

This job answers while someone waits, so latency comes first: we hold it to a single small card and, within that, take the best independent score on structured output and tool calls.

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.

GLM-4.672.38%Kimi K259.06%DeepSeek-V3.254.12%Qwen3-32B48.71%Qwen3-235B-A22B47.99%Qwen3-8B42.57%Qwen3-30B-A3B41.39%Qwen3-14B41.03%
Longer is betterFree to serveConditions apply⚠ answered under 95%

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

Sticking to the document

Vectara Hallucination Leaderboard · HHEM · 17 of 18 models measured

Measured on public documents, by a model acting as judge. Read it beside the answer rate: the lowest hallucination rates on this board belong to models that simply decline more often.

Phi-43.7% · answers 80.7% ⚠Llama 3.3 70B4.1% · answers 99.5%Gemma 3 12B4.4% · answers 97.4%Qwen3-8B4.8% · answers 99.9%Mistral Small 3.25.1% · answers 97.9%Granite 4.0 Small5.2% · answers 100%DeepSeek-V3.25.3% · answers 96.6%Qwen3-14B5.4% · answers 99.9%
Shorter is betterFree to serveConditions apply⚠ answered under 95%

Independent measurement · Vectara · board updated May 11, 2026

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.

  1. 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.

  2. 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.

  3. 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 customer service

All 15 →

Next

Tell us what your team does by hand

Describe the process that takes the most time. We will say plainly whether a model is the right tool for it.