A model you own
from your traffic.

Turn on training and your traffic becomes a model for your work. It serves what it passes on your evals and the weights are yours.

Traces from your traffic

123

Level 3

Your model

A model trained on your work that serves what it passes. Your frontier model stays as fallback.

- model: "gpt-6-sol"
+ model: "your-model-v1"

Scored against frontier models. Both studies are public.

Claims research · August 2026

A trained Qwen3.5-4B on 500 held-out claims from the public AVeriTeC benchmark.

Computer use · July 2026

Qwen 3.5 9B distilled from Kimi K2.6 working 1,000 tasks on Mac VMs.

The claims study put a small router in front of three models. Each point is measured on 489 claims the router never saw.

Verdict accuracy against cost per 1,000 claims

1
Turn it on
Training is opt-in. Nothing trains on your traffic until you switch it on.
2
We simulate
Your traces become a simulation of the workload with your tools and your data shapes.
3
We train
Distillation from a frontier teacher or reinforcement learning in the simulation. The signal in your work picks the method.
4
You approve
The candidate shadows real traffic first. You move the alias when the numbers clear your bar.

client.chat.completions.create({
-  model: "gpt-6-sol",
+  model: "your-model-v1",
   messages,
})

A managed training engagement. If the model ships the fee goes toward its inference.