Tracking and Routing LLM Prices to Cut Costs

Forsaken-Bobcat4065 · reddit · 2026-07-13

The author shares a script that checks LLM prices every 3 hours to catch unannounced hikes or drops early, preventing end-of-month bill shocks.

Key takeaways from the post:

After a price drop for GLM-5.2, it costs about $1.20 / M tokens for input and $4.10 / M tokens for output, which the author estimates to be roughly 1/6 the cost of GPT-5.5.
It offers great cost-efficiency for batch code, document, and unit test generation. However, full deep reasoning is somewhat slow, making it better suited for real-time chat with reasoningeffort turned down.
By routing complex architecture reviews to Claude and standard batch tasks to cheaper models, the author cut monthly token spend by about 30%.
The post also includes a routing example based on OpenAI-compatible APIs and a price-tracking method using scheduled JSON fetches via the /models endpoint and diffing the results.

More from Venture