The patch counts idle-period usage against team shares, moving the light team from 13 of 20 requests to 20 of 20 in a live Anthropic test.
By [RuntimeWire Staff](https://runtimewire.com/author/runtimewire-staff)
· Published
LiteLLM, the AI gateway co-founded by Krrish Dholakia and Ishaan Jaffer, has merged a dynamic rate limiter fix after a test in which a busy team consumed 901,543 tokens on a model where two teams were each owed 500,000. Before the patch, the lighter team in the same test stopped at 130,359 tokens and received seven model-capacity 429 responses. After the patch, the busy team stopped at 560,948 tokens and the light team completed all 20 of its requests.…