SLM Arena

Fine-tuned models. Real benchmarks. No hype.

Every model in this arena was trained on real domain data, evaluated against specialist ground truth, and has a published hallucination rate. Request access or join the waitlist for any model.

Request model access Compare benchmarks

ARENA STATS
6
Models available
3
Verticals covered
1B–7B
Parameter range
<1%
Avg hallucination rate

Benchmark comparison

All models. All numbers.

Every figure is measured on held-out test sets. Accuracy is against specialist-annotated ground truth. Hallucination rate is the fraction of outputs containing factually incorrect statements.

Model	Vertical	Params	Accuracy	Latency p50	Hallucination rate	Context	Access

How to access

Three steps to production.

Request access

Fill in the contact form with your use case. We review every request within 24 hours and send API credentials + documentation.

Run your benchmark

We provide a benchmark harness and sample datasets. You run it on your own test set and get an accuracy report before committing to production.

Deploy on your infra

We ship container artifacts (.tar.gz + signatures). You deploy to your cluster. We provide a 30-day deployment SLA and a signed inference manifest from day one.