---
type: "post"
id: "liquid-ai-launched-ripl-a-sub-50ms-inference-10eb44"
slug: "liquid-ai-launched-ripl-a-sub-50ms-inference-10eb44"
title: "Liquid AI launched RIPL , a sub-50ms inference engine for LFM2.5 models on NVIDIA GPUs."
description: "Liquid AI has announced RIPL, short for Real-time Inference Pipeline, an inference engine for its LFM2.5 models."
retrieval_nugget: "Liquid AI has announced RIPL, short for Real-time Inference Pipeline, an inference engine for its LFM2.5 models."
published_at: "2026-10-23"
updated_at: "2026-10-23"
record_date: "2026-10-23"
date_kind: "published_at"
topics: ["inference"]
entities: ["Liquid AI","RIPL","LFM2.5","NVIDIA","Shopify","vLLM"]
related_patterns: []
novelty_label: "new_fact"
what_changed: "Liquid AI introduces RIPL, a new inference engine for LFM2.5 with sub-50 ms end-to-end latency on NVIDIA GPUs; the prior covered DSpark speculative decoding for faster LFM2.5 inference."
prior_refs: ["memory:cd81fb1c-b5d7-8bf1-a387-6c0ca4fe4a8b"]
source_url: "https://www.liquid.ai/blog/low-latency"
source_urls: ["https://www.liquid.ai/blog/low-latency"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:liquid-ai-launched-ripl-a-sub-50ms-inference-10eb44"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/liquid-ai-launched-ripl-a-sub-50ms-inference-10eb44/","markdown":"https://newruntime.com/posts/liquid-ai-launched-ripl-a-sub-50ms-inference-10eb44.md","json":"https://newruntime.com/posts/liquid-ai-launched-ripl-a-sub-50ms-inference-10eb44.json"}
---

# Liquid AI launched RIPL , a sub-50ms inference engine for LFM2.5 models on NVIDIA GPUs.

## Retrieval answer

Liquid AI has announced RIPL, short for Real-time Inference Pipeline, an inference engine for its LFM2.5 models.

Liquid AI has announced RIPL, short for Real-time Inference Pipeline, an inference engine for its LFM2.5 models. The company says the engine is designed for live request paths with tight latency limits and can return complete responses in under 50 ms under continuous peak traffic.

The company reports that LFM2.5-1.2B on one NVIDIA B200 reached 34.56 ms p95 latency at 1,000 requests per second for a 32-token input and 32-token output workload. LFM2.5-2.6B recorded 76.44 ms in the same test. Liquid also reports a 24.2 ms end-to-end p95 result at 500 requests per second in a recorded latency trace.

RIPL is available now for enterprise deployments of Liquid Foundation Models on NVIDIA H100, H200, B200, and RTX PRO 6000 GPUs. Liquid cites Shopify as deploying LFMs for latency-sensitive commerce workloads, including search enhancement.

The published comparisons do not establish an apples-to-apples model benchmark: LFM2.5 runs on RIPL while MiniCPM, Granite, and Qwen models run on vLLM. The source does not state RIPL pricing, general self-service availability, its technical architecture, or results from an independent evaluation.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Inference](https://newruntime.com/topics/inference.md): Explore the inference topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)