Ticket routing reads an incoming message — email, web form, chat transcript — and sends it to the queue that should handle it, with a priority attached, so that a customer is not waiting while their message is passed between teams.
A routing decision is several judgements at once: which product or department the message concerns, how urgent it is, which language it is in, and whether it resembles an open ticket already in the system. Each can be a small model, and running them together costs little. What determines success is less the model than the handling of uncertainty: a confident decision routes automatically, an unconfident one goes to a person, and both are recorded so the model can be measured against what the team actually did.
Models in this group take text or whole documents as input: plain text, PDFs, scanned pages and office files. Each specification table gives three hardware tiers — Minimum, the smallest setup on which the model runs correctly; Medium, the usual production configuration; and High, a configuration sized for peak volume. The rate is the number of sample inputs processed per hour on that hardware. Use these rates for initial sizing. Before production, benchmark your own data to validate accuracy, latency, throughput and cost. The sample input here is one ticket of about 200 words routed to a queue with a priority attached.
DeBERTa v3
Vendor: Microsoft
What it does: fitted to your queues from your own ticket history, so it learns the routing your team already practises rather than a generic taxonomy. The most accurate option where history exists, and cheap enough to run on every ticket.
Requirement
Minimum
Medium
High
GPU type
RTX 3060
RTX 4090
1× A100 80 GB
VRAM
6 GB
24 GB
80 GB
vCPUs
4
8
24
RAM
16 GB
32 GB
64 GB
Server
1× RTX 3060 12 GB
1× RTX 4090 24 GB
1× A100 SXM 80 GB
Rate (tickets/hour)
≈ 30,000
≈ 120,000
≈ 340,000
SetFit
Vendor: Hugging Face
What it does: stands up a route for a brand-new queue from a handful of example tickets, so a product launch or an incident queue is covered immediately instead of after weeks of data collection.
Requirement
Minimum
Medium
High
GPU type
RTX 3060
RTX 4090
1× A100 80 GB
VRAM
6 GB
24 GB
80 GB
vCPUs
4
8
24
RAM
16 GB
32 GB
64 GB
Server
1× RTX 3060 12 GB
1× RTX 4090 24 GB
1× A100 SXM 80 GB
Rate (tickets/hour)
≈ 28,000
≈ 110,000
≈ 320,000
XLM-RoBERTa
Vendor: Meta
What it does: detects the language of a ticket and routes it across a hundred languages with a single model, which keeps a multilingual desk on one pipeline.
Requirement
Minimum
Medium
High
GPU type
RTX 3060
RTX 4090
1× A100 80 GB
VRAM
8 GB
24 GB
80 GB
vCPUs
6
8
24
RAM
16 GB
32 GB
64 GB
Server
1× RTX 3060 12 GB
1× RTX 4090 24 GB
1× A100 SXM 80 GB
Rate (tickets/hour)
≈ 26,000
≈ 100,000
≈ 300,000
BGE-M3
Vendor: Beijing Academy of Artificial Intelligence
What it does: turns each ticket into a numeric fingerprint, which finds tickets already in the system describing the same problem. That both prevents duplicate work and surfaces the resolution that worked last time.
Requirement
Minimum
Medium
High
GPU type
RTX 3060
RTX 4090
1× A100 80 GB
VRAM
8 GB
24 GB
80 GB
vCPUs
6
12
24
RAM
16 GB
32 GB
64 GB
Server
1× RTX 3060 12 GB
1× RTX 4090 24 GB
1× A100 SXM 80 GB
Rate (tickets/hour)
≈ 15,000
≈ 70,000
≈ 200,000
Llama 3.1 8B
Vendor: Meta
What it does: routes from written queue descriptions, sets a priority, and states in one sentence why it chose that queue — which is what lets a supervisor audit the routing and correct the rules.
Requirement
Minimum
Medium
High
GPU type
RTX 3090
RTX 4090
1× H100 80 GB
VRAM
16 GB
24 GB
80 GB
vCPUs
8
12
24
RAM
32 GB
48 GB
128 GB
Server
1× RTX 3090 24 GB
1× RTX 4090 24 GB
1× H100 SXM 80 GB
Rate (tickets/hour)
≈ 2,000
≈ 6,000
≈ 20,000
Mistral Small 3
Vendor: Mistral AI
What it does: handles long chat transcripts and email threads where the actual request appears halfway down, and can draft the first reply at the same time as routing.
Requirement
Minimum
Medium
High
GPU type
RTX 4090
1× L40S 48 GB
1× H100 80 GB
VRAM
24 GB
48 GB
80 GB
vCPUs
12
16
32
RAM
48 GB
64 GB
128 GB
Server
1× RTX 4090 24 GB
1× L40S 48 GB
1× H100 SXM 80 GB
Rate (tickets/hour)
≈ 1,300
≈ 3,800
≈ 11,000
Choosing between them
Routing accuracy depends on your queue structure and how consistently your own team has routed in the past. Our consultants review your ticket history and queue definitions, then recommend the models, the confidence thresholds for automatic routing, and the reporting that shows what the system is getting right.
At the start of a project we may run a short proof of concept on a sample of your own data, measuring the accuracy and the throughput the model actually achieves on your material. That replaces the estimates on this page with real figures, so the cost and the schedule for the full engagement are known before it is committed.
Share a representative sample, expected volume, latency target and deployment location for Customer-support ticket routing. We benchmark the shortlisted models, recommend the lowest-cost GPU configuration that meets the target, and scale it from pilot capacity to a dedicated production cluster.