2.61Btokens through the router in 14 days$23.85total model spend at list price29.9%of requests failed
Fourteen days of router logs from a research workload running on nine free or near-free LLM providers: 83,497 requests, 2.6 billion tokens, a 29.9% failure rate, and the eleven changes that made it usable.
ai-infrastructurellm-routingfree-modelsresearch-automation