Prime Inference reports 96.9% cache hit rate for Nemotron 3.5 Super VL

Prime Intellect says its Prime Inference service has a 96.9% cache hit rate for Nemotron 3.5 Super VL, the highest of any provider serving that model.

Prime Intellect says the rate brings Prime Inference's effective input cost to $0.024 per million tokens, the lowest of any provider and about 70% below the largest provider. The claim comes from a Prime Intellect post on X dated November 2, 2026, and no independent figures are reported.