Batch inference, at the batch rate
polyrouter now runs batch jobs on the upstream batch API, at the batch rate. Getting there meant reclassifying 69 OpenRouter rows from models into prices.
Read it →Engineering write-ups and practical notes on routing LLM traffic — what polyrouter decides, how it degrades, and what it costs. Roughly twice a month. RSS.
polyrouter now runs batch jobs on the upstream batch API, at the batch rate. Getting there meant reclassifying 69 OpenRouter rows from models into prices.
Read it →Manifest deprecated rule-based routing. polyrouter kept automatic routing and hardened it. What a self-hosted LLM router adds, and how the two compare, in code.
Read it →