Comparison

LibreQoS and Bequant

Both deploy inline at Layer 2. The similarity ends there. Bequant controls traffic with TCP proxies, payload inspection, and a proprietary congestion algorithm it markets as AI. LibreQoS runs CAKE and fq_codel — published, peer-reviewed AQM that has been the reference answer to bufferbloat since 2012, shipping in the mainline Linux kernel. Capabilities below are cited from each side's own materials.

Queue Discipline You Can Verify

Why the marketing label matters

Bequant markets its congestion handling as AI-driven. Whatever the algorithm does, it cannot be inspected: there is no published research, no source, and no way to know why a subscriber was throttled at busy hour.

LibreQoS takes the opposite trade. Its queue discipline is documented in academic literature and readable in source, and its guardrails (TreeGuard, observe mode) are deterministic and yours to configure. When something misbehaves, you can find out why.

Comparison

Platform by platform, with sources

Primary mechanism
LibreQoS
Per-flow AQM and fair queueing: CAKE or fq_codel leaf qdiscs inside HTB classes for every circuit, site, AP, and OLT LibreQoS docs
Bequant
TCP proxying and optimization, plus a proprietary congestion-management algorithm marketed as AI-based Bequant docs
Protocol coverage
LibreQoS
All flows. Classification is an IP 5-tuple hash, so TCP, QUIC, and UDP receive the same queue discipline LibreQoS docs
Bequant
TCP optimization accelerates TCP; DPI classifies applications for policy Bequant TCPO docs
Application DPI
LibreQoS
Not required. Flow identity resolves through a flow cache and longest-prefix match on addressing data LibreQoS docs
Bequant
DPI identifies application categories for app-aware policies and per-application metrics Bequant policies docs
Shared capacity
LibreQoS
Explicit hierarchy: Site, AP, OLT, and backhaul capacity each get their own HTB class LibreQoS docs
Bequant
Inline across 1 to 32 links, deployed from the Internet uplink to the network edge Bequant deployment docs
Deployment
LibreQoS
Layer-2 Linux bridge, inline by default, with a documented bypass path. Single-interface mode is advanced LibreQoS bridge guide
Bequant
Layer 2 inline on standard Intel or AMD servers, including VMware and KVM Bequant deployment docs
Throughput ceiling (published)
LibreQoS
Documented reference builds to 100 Gbps, with published per-CPU capacity tables and multi-node hierarchy beyond a single box LibreQoS sizing
Bequant
Up to 400 Gbps per server, with clustering across servers Bequant platform
Subscriber rates and billing
LibreQoS
Per-circuit min/max rates synced from Splynx, UISP, Sonar, Powercode, VISP, WISPGate, Netzur, or RADIUS dynamic circuits LibreQoS integrations
Bequant
Subscriber plans imported via RADIUS, billing integrations, or REST API Bequant billing docs
Licensing model
LibreQoS
Subscriber-based subscription. No throughput licensing on the shaping core LibreQoS pricing
Bequant
Per-server monthly subscription based on peak throughput processed Bequant platform
Source
LibreQoS
GPL2 open-source core; Insight and the API package are commercial GitHub
Bequant
Proprietary platform Bequant platform
Inspectability
LibreQoS
GPL2 source, published queue behavior, and deterministic guardrails you can read and configure GitHub
Bequant
Proprietary; the congestion algorithm cannot be audited Bequant platform
QoE evidence
LibreQoS
Per-circuit RTT, retransmits, flows, and QoO, with Insight heatmaps and ASN analysis on top Insight
Bequant
Per-subscriber, per-application, and network metrics measured on traffic the BQN proxies and shapes Bequant visibility docs
Independent verification
LibreQoS
Public Internet Quality Test anyone can run, plus observe mode for honest before/after baselines Quality test
Bequant
Free trial with deployment support and trial-result analysis Bequant platform

Bequant capabilities are summarized from Bequant's published platform materials and user manual; claims that cannot be independently verified are identified as marketing claims, not verified capabilities. LibreQoS capabilities are summarized from the LibreQoS documentation and pricing page.

The Real Difference

A speed boost for some traffic, or predictable behavior for all of it

A TCP proxy makes recognized TCP flows faster. Its benefit is conditional: on the flow being TCP, on the proxy seeing it in both directions, and on the patterns holding as applications move to encrypted QUIC and UDP. The queue underneath still balloons for everything the proxy does not touch.

Published AQM is unconditional. CoDel was published by Nichols and Jacobson in 2012; fq_codel and CAKE ship in the mainline Linux kernel; the IETF community has spent a decade documenting bufferbloat as a queueing problem. Active queue management keeps latency low at the point of congestion for every flow, and the hierarchy keeps local congestion from spreading upstream. Nothing depends on recognizing the application.

There is a licensing consequence too. When capacity is licensed per server against peak throughput, every traffic growth event is a license event. Subscriber-based pricing tracks the size of your business instead.

One side of this comparison asks for trust; the other offers evidence. Published AQM does not need a vendor word to be believed — its behavior is in the literature, in the kernel, and measurable on your own network with a public test. Trust is not a deployment requirement here.

Ask Both Vendors

Questions worth asking in any Bequant evaluation

  1. 01What happens to QUIC and UDP flows that the TCP optimizer does not accelerate?
  2. 02Which decisions rely on DPI, and what is the privacy posture for payload inspection in our jurisdiction?
  3. 03How is congestion at a specific AP or OLT contained, rather than managed at the uplink where the appliance sits?
  4. 04What does the subscription do at the next throughput milestone?
  5. 05Can we measure the before/after ourselves, with a public test and an unshaped baseline, before committing?

Verify It Yourself

Don't decide on vendor claims

Run the public Internet Quality Test on a busy-hour circuit, deploy LibreQoS on spare hardware or a pilot site, and run it again. Idle and loaded latency, loss, throughput, and a QoO grade — measured on your own network.

Keep Comparing

Other comparisons

Deploy

Evaluate LibreQoS against your requirements

Run the 1-line install on a pilot site, point your topology at it, and measure busy-hour latency before and after.