Liquid AI releases RIPL, a real-time inference engine for LFM models

Liquid AI has released RIPL, which it describes as a real-time inference engine for its LFM models. The launch is framed around low-latency serving of those models.

In the benchmark reported with the release, LFM2.5-1.2B running on RIPL on a single B200 showed 14.4 to 28.4 times lower p95 latency than Qwen3.5-0.8B running on vLLM.

Developers running small LFM models behind agent tool calls can compare p95 latency against their current serving stack before changing anything.

The figures come from one B200 and one model pairing against vLLM, so they do not show how RIPL performs on other hardware, model sizes or workloads.