Where each frontier is positioned, per Glean's scoring:

> GPT-5.6 Luna (xhigh) at 55, $0.08 per task
> GLM 5.2 (high) at 57, $0.35
> Gemini 3.7 Flash (high) at 61, $0.47
> Kimi K3 (high) at 63, $0.90
> Claude Opus 5 (high) at 67, $2.96

Glean’s own autorouting works by using Glean Waldo, a small specialist model post-trained on NVIDIA Nemotron 3 Nano, to set the reasoning level at runtime, then hand off to a specialist model when the task needs it.

> Model selection across 40+ open and frontier models.

> Effort routing sets how hard the model thinks.

Check it out! 🔥