LMCache vs sizeless: Features, Pricing & Which Is Better (2026)
A side-by-side comparison of LMCache and sizeless — features, pricing, and ideal use cases — to help you decide which AI tool fits your workflow.
L
LMCache
LMCache
LMCache is an open-source KV cache layer that speeds up LLM inference by storing and reusing KV caches across GPU, CPU, disk, and S3.
Key features
- KV Cache Reuse: Stores KV caches of reusable text across the datacenter so prefixes are not recomputed across requests or serving engines.
- Multi-Tier Storage: Persists caches across GPU, CPU, local disk, and S3 with acceleration techniques like zero CPU copy, NIXL, and GDS.
- vLLM Integration: Combines with vLLM to deliver 3-10x reductions in delay and GPU cycles for multi-round QA and RAG workloads.
- Pluggable KV Transformation: A flexible SERDE interface lets researchers add compression, token dropping, and custom serialization.
- Vendor-Neutral Layer: Works as a KV cache layer across mainstream serving engines, inference frameworks, hardware vendors, and storage systems.
- Faster Time-to-First-Token: Cuts TTFT and improves throughput for long-context, agentic, and knowledge-augmented workloads.
Best for
- Retrieval-Augmented Generation: Reuse cached document prefixes to cut latency and GPU cost in RAG pipelines.
- Multi-Turn Conversations: Avoid recomputing conversation-history KV caches across turns in chat applications.
- Long-Context Agents: Accelerate agentic workloads that repeatedly process large shared context.
- Enterprise-Scale Inference: Share KV caches across multiple serving instances to raise throughput in production clusters.
- Cache Compression Research: Prototype custom KV compression and serialization through the pluggable SERDE interface.
sizeless
sizeless
Turns a smartphone video of an open trench into a centimetre-accurate 3D point cloud, CAD as-built plan and GIS-ready digital twin of buried utilities.
Key features
- Smartphone capture: Field crews record an open trench with a standard iPhone Pro — no specialist scanning hardware and no separate surveying appointment
- Centimetre-accurate point clouds: Reconstruction algorithms developed at ETH Zurich build a high-resolution 3D point cloud of the excavation from the video alone
- Standards-compliant CAD output: Generates as-built plans in DWG and DXF, with couplings and pipe runs identified and measurements simplified
- 3D digital twin and GIS export: Produces a model of the pipe route including building entries that drops into existing GIS systems
- Works without GPS: Captures basement sections and building entry points where GNSS-based surveying fails
- Immediate backfilling: Because capture takes minutes, trenches close right after filming instead of waiting on a survey crew
- Documentation in about 72 hours: Complete records arrive weeks earlier than conventional surveying, enabling prompt connection billing
- Third-party utility capture: Records crossing utilities and as-laid geometry as unbroken 3D evidence, replacing hand sketches
Best for
- A utility network operator documenting residential service connections without booking a surveyor for every site
- A contractor closing a trench the same day instead of leaving it open pending a survey appointment
- Capturing a building entry point in a basement where GPS-based surveying cannot get a fix
- A district heating project producing as-built DWG plans for regulatory sign-off
- Spotting a laying error in the 3D point cloud before backfilling, while the fix is still cheap
- Feeding as-built pipe geometry into a GIS system for long-term network maintenance planning