Best software load balancers for bare-metal Kubernetes clusters

Combined ranking

1. 1
   MMetalLBGrade ↗—14 pts

   GPT #1Claude #2Gemini #1

   Best general-purpose choice: a mature, CNI-agnostic LoadBalancer implementation with simple ARP/NDP mode and production BGP/ECMP through FRR-K8s, including BFD, dual-stack, IPv6, and routing-policy controls.

   + model takes & fixes− hide details

   GPT Best general-purpose choice: a mature, CNI-agnostic LoadBalancer implementation with simple ARP/NDP mode and production BGP/ECMP through FRR-K8s, including BFD, dual-stack, IPv6, and routing-policy controls.

   Gemini Battle-tested open-source standard for bare-metal Service type=LoadBalancer; provides simple Layer 2 (ARP/NDP) mode for basic clusters and robust BGP mode (via FRR) for ECMP active-active routing with physical network switches. Flagged near-tie with Cilium, earned top spot due to universal CNI-agnostic deployment flexibility.

   Claude the de facto standard bolt-on for bare-metal LoadBalancer services — CNI-agnostic, tiny footprint, well-documented, with both L2 (ARP/NDP failover) and BGP modes and CRD-based IP pools; the safe default for a typical practitioner who just needs external IPs and doesn't want to change their networking stack

   Where it falls short

   per GPT It supplies IP allocation and advertisement, not an L7 proxy or complete dataplane; performance and traffic behavior still depend on kube-proxy/CNI and the physical network.

   per Claude L2 mode funnels all traffic for a service through one elected node (no true load spreading, failover-only) and BGP mode leans on a single-path model with limited ECMP nuance; it's a plain address advertiser, not a full-featured LB with health-based traffic shaping

   per Gemini Layer 2 mode forces all traffic for a single VIP through a single leader node creating bottleneck vulnerabilities, while BGP mode requires external network router configuration and network engineering access.

2. 2
   CiliumGrade ↗Visit ↗incumbent—13 pts

   GPT #2Claude #1Gemini #2

   eBPF datapath doubles as kube-proxy replacement, so a single CNI both routes and advertises LoadBalancer IPs — BGP peering for routed environments, L2/ARP announcements for flat networks, plus LB-IPAM pools, DSR, Maglev consistent hashing, and observability via Hubble; the strongest choice when you're already committing to Cilium as the CNI and want L4 service LB without a second component

   + model takes & fixes− hide details

   Claude eBPF datapath doubles as kube-proxy replacement, so a single CNI both routes and advertises LoadBalancer IPs — BGP peering for routed environments, L2/ARP announcements for flat networks, plus LB-IPAM pools, DSR, Maglev consistent hashing, and observability via Hubble; the strongest choice when you're already committing to Cilium as the CNI and want L4 service LB without a second component

   GPT Strongest integrated option when Cilium is already the CNI: eBPF service handling, LB IPAM, BGP/ECMP, Maglev, DSR, dual-stack, and Hubble observability can replace several networking components while delivering excellent performance.

   Gemini eBPF-native L4 load balancing with LB-IPAM and integrated BGP control plane that bypasses kube-proxy completely, delivering unmatched packet throughput, Maglev consistent hashing, and Hubble network visibility. Flagged near-tie with MetalLB for modern greenfield clusters.

   Where it falls short

   per GPT Adopting it principally for load balancing is a major networking commitment, and its L2-announcement API remains beta.

   per Claude value is tied to running Cilium itself — if your cluster is on Calico/Flannel or you just want a bolt-on LB, adopting the whole CNI to get service load balancing is disproportionate, and the eBPF/BGP surface has a real learning curve

   per Gemini Tight coupling that requires Cilium as the cluster CNI and modern Linux kernel support, introducing unnecessary operational complexity for teams only needing a standalone load balancer.

3. 3
   Kkube-vipGrade ↗—8 pts

   GPT #4Claude #3Gemini #3

   uniquely covers both the control-plane VIP (HA API server without an external LB) and service LoadBalancer IPs in one lightweight daemon, via ARP (leader-elected) or BGP; ideal for edge, on-prem, and self-managed clusters that need HA masters and service LBs from a single tool

   + model takes & fixes− hide details

   Claude uniquely covers both the control-plane VIP (HA API server without an external LB) and service LoadBalancer IPs in one lightweight daemon, via ARP (leader-elected) or BGP; ideal for edge, on-prem, and self-managed clusters that need HA masters and service LBs from a single tool

   Gemini Unifies bare-metal Kubernetes API server control-plane high availability and workload Service type=LoadBalancer IP allocation into a single, lightweight DaemonSet or static pod architecture using ARP, NDP, and BGP without external controller dependencies.

   GPT Lightweight and actively maintained, with ARP, BGP, dual-stack, per-service leader election, DHCP/VLAN support, and the unusual value of handling both Kubernetes control-plane VIPs and LoadBalancer Services.

   Where it falls short

   per GPT Service IP allocation commonly requires its separate cloud-controller, while ARP mode places each VIP’s traffic through one elected node; near-tied with PureLB, it ranks higher for maturity and simplicity.

   per Claude service-LB feature depth and IPAM ergonomics trail MetalLB/Cilium, and in ARP mode it shares the same single-node-per-service bottleneck; its sweet spot is control-plane HA more than heavy north-south service traffic

   per Gemini Lacks high-throughput eBPF packet processing, Maglev load distribution, and granular traffic management features required for multi-terabit bare-metal edge environments.

4. 4
   PPureLBGrade ↗—4 pts

   GPT #5Claude #4Gemini #5

   a MetalLB-lineage service LB built around Linux routing — its "virtual network" model plays cleanly with any external router or routing daemon (FRR, BIRD), giving more flexible BGP/ECMP integration and multi-interface setups than MetalLB for network-savvy teams

   + model takes & fixes− hide details

   Claude a MetalLB-lineage service LB built around Linux routing — its "virtual network" model plays cleanly with any external router or routing daemon (FRR, BIRD), giving more flexible BGP/ECMP integration and multi-interface setups than MetalLB for network-savvy teams

   GPT A capable Linux-native design with local and routed address pools, dual-stack, GoBGP/ECMP, Direct Server Return for local-endpoint services, multi-subnet support, and unusually good operational inspection.

   Gemini Innovative architecture that assigns LoadBalancer IPs directly to native Linux virtual network interfaces (macvlan/dummy) and integrates with standard host routing daemons, avoiding ARP spoofing hacks and custom BGP controllers.

   Where it falls short

   per GPT Its newer implementation and substantially smaller deployment community make it less proven operationally than the leaders; near-tied with kube-vip for advanced on-prem routing use.

   per Claude much smaller community and thinner documentation than MetalLB or Cilium, so you're more on your own operationally; the added flexibility only pays off if you actually run and understand external routing infrastructure

   per Gemini Significantly smaller community adoption, ecosystem documentation, and vendor support footprint compared to MetalLB and Kube-VIP, with complex configuration requirements on multi-homed server nodes.

5. 5
   CCalicoGrade ↗—3 pts

   GPT #3Claude —Gemini —

   Excellent for BGP-based bare-metal networks, combining mature routing, LoadBalancer IPAM, ECMP service advertisement, source-IP preservation, and optional eBPF DSR/Maglev in one established CNI.

   + model takes & fixes− hide details

   GPT Excellent for BGP-based bare-metal networks, combining mature routing, LoadBalancer IPAM, ECMP service advertisement, source-IP preservation, and optional eBPF DSR/Maglev in one established CNI.

   Where it falls short

   per GPT Its advantages are strongest only when the cluster already uses Calico and the upstream network supports BGP; it is not the easiest drop-in L2 solution.

6. 6
   HHAProxy EnterpriseGrade ↗—2 pts

   GPT —Claude —Gemini #4

   Unrivaled raw software L4/L7 proxy performance, extreme process efficiency, advanced health checking, and enterprise-grade security for edge ingress routing on bare-metal hardware.

   + model takes & fixes− hide details

   Gemini Unrivaled raw software L4/L7 proxy performance, extreme process efficiency, advanced health checking, and enterprise-grade security for edge ingress routing on bare-metal hardware.

   Where it falls short

   per Gemini Not a native Kubernetes LoadBalancer IPAM controller out of the box, requiring integration with Keepalived/VRRP or Ingress wrappers to handle dynamic Service IP allocations.

7. 7
   FF5 BIG-IPGrade ↗—1 pts

   GPT —Claude #5Gemini —

   the strongest commercial option — offloads L4/L7 to hardware or virtual BIG-IP appliances with mature health monitoring, WAF, SSL offload, and vendor support, integrated into Kubernetes via CIS/Container Connector; the pick for regulated enterprises that already own F5 and need supported, audited traffic infrastructure

   + model takes & fixes− hide details

   Claude the strongest commercial option — offloads L4/L7 to hardware or virtual BIG-IP appliances with mature health monitoring, WAF, SSL offload, and vendor support, integrated into Kubernetes via CIS/Container Connector; the pick for regulated enterprises that already own F5 and need supported, audited traffic infrastructure

   Where it falls short

   per Claude licensing cost, appliance dependency, and operational heaviness make it wrong for anyone wanting a self-contained software LB; you're buying an external device ecosystem, not a cloud-native in-cluster component

Just missed the top 5

By model

Common questions

Tracked by ModelsAgree · rank 1 = 5 pts … rank 5 = 1 pt · re-polled on demand