Blog

Notes from the routing layer.

Engineering, model evaluation, pricing tear-downs. Posts ship weekly once we open the gates. RSS feed will appear at this URL once the first post lands.

First three posts in the queue

DeepSeek V3.2 vs Claude Opus 4.7 — head-to-head on real coding agents

Draft

We ran 200 real Cursor sessions through both and measured: time-to-first-PR, edit-over-edit success rate, and total dollars spent. Spoiler: DeepSeek wins on $/edit at 18× lower cost.

Planned: Week of 2026-05-19

Why Kimi K2.5 quietly became the long-context champion

Draft

Anthropic's 200K is the floor, but Kimi's 128K + tool-use loop is what production agents reach for. A teardown of the routing patterns that emerged in our first 1M requests.

Planned: Week of 2026-05-26

OpenAI / Anthropic / Google rate-card archaeology

Draft

Every public price change since 2024, charted, with the Routify pass-through that follows. Public spreadsheet linked.

Planned: Week of 2026-06-02

Want the first post in your inbox?

Drop your email — we'll only ship when we have something engineering-worthy to say.

Join waitlist