Three of fourteen models named HAProxy first on the direct prompt; one named Azure Load Balancer. HAProxy was named by fourteen of the fourteen models and Azure Load Balancer by thirteen and HAProxy carries 58 labels and Azure Load Balancer 29, so the shares are not directly comparable.
Named in two categories this edition.
Named in one category this edition.
Share is the count of first choices across the direct, paraphrase, budget and scale prompts over all fourteen models, for a mid-market B2B company; rank is within the category; every quote names the model and the prompt it came from. Both figures come from the load balancers page.
Across every category in the October 2026 Edition, HAProxy and Azure Load Balancer were named in the same answer sixty-five times, of the 174 answers naming HAProxy and the 83 naming Azure Load Balancer. In those answers Azure Load Balancer took the first choice four times and HAProxy nineteen.
| Model | Direct | Paraphrase | Comparative | Budget-constrained | Scale-constrained | Negative |
|---|---|---|---|---|---|---|
| Claude Haiku 4.5 | ||||||
| GPT-5.4 mini | ||||||
| Gemini 3.5 Flash | ||||||
| Perplexity Sonar | ||||||
| Grok 4.1 Fast | ||||||
| Mistral Small | AL | AL | ||||
| DeepSeek V4 Flash | ||||||
| Llama 4 Maverick | ||||||
| Qwen 3.7 Flash | ||||||
| Kimi K2 | ||||||
| GLM 4.7 FlashX | ||||||
| MiniMax M2.5 | AL | AL | ||||
| GPT-6 Luna | AL | |||||
| Muse Glimmer 30B |
Bold names in an answer are the products the judge labeled a first choice; a model naming several gives each of them that label. The full answer text for every row is in the record.
Every negative label with a quote, up to three, then the highest-weighted positives, up to three. Six of seven in this category shown.
“Open‑source solutions that are no longer actively maintained (e.g., very old versions of HAProxy, Nginx...) | Unpatched CVEs” GLM 4.7 FlashX · negative prompt · soft negative
“consider a third-party option like F5 or HAProxy, but that adds more ops overhead” GPT-5.4 mini · direct prompt · soft negative
“running them on under-powered virtual machines (like a $5/mo VPS) creates a bottleneck” Qwen 3.7 Flash · negative prompt · soft negative
“If you have standard web apps or APIs: Go with HAProxy. It has the broadest adoption, most documentation, and best performance for the price (free).” Kimi K2 · budget prompt · first choice
“The performance leader — handles millions of requests per second with sub-millisecond latency... HAProxy is the performance benchmark.” DeepSeek V4 Flash · comparative prompt · first choice
“Fastest and most stable - used by GitHub, Reddit, Stack Overflow ... High-performance, mission-critical | HAProxy” Kimi K2 · comparative prompt · first choice
Every negative label with a quote, up to three, then the highest-weighted positives, up to three. Six of eight in this category shown.
“Cloud-native load balancers (AWS ELB, Azure LB, GCP LB) ... may lack deep packet inspection, advanced routing rules, or cross-cloud consistency” MiniMax M2.5 · negative prompt · soft negative
“Traefik is rated higher for value, with reviewers expressing greater satisfaction with its pricing compared to Azure Load Balancer.” Muse Glimmer 30B · budget prompt · soft negative
“excellent within their ecosystems but can lead to vendor lock-in” Mistral Small · negative prompt · soft negative
“Cloud-native load balancers (AWS ALB, Azure LB) are often the best fit” Claude Haiku 4.5 · direct prompt · first choice
“Azure Load Balancer for network-layer distribution, especially in Azure-native environments.” GPT-5.4 mini · direct prompt · alternative
“AWS Elastic Load Balancing, Azure Load Balancer, or Google Cloud Load Balancing” MiniMax M2.5 · budget prompt · alternative
Comparisons are drawn for the top eight products in each category, each against each. The output is the models' output; nothing here is a recommendation by the index.