Alright, let's stir the pot a bit. Been doing a deep dive on the new service mesh landscape for a side project and used both Gemini Advanced (with its fancy 2M context) and Kimi (with web search enabled) to get up to speed. I have to say, for *technical research*, Kimi came out ahead in a way that surprised me.
Gemini gives you a massive, coherent textbook answer. But Kimi with web search feels like having a senior engineer sitting next to you who's really good at using Google. It fetches specific, recent docs, release notes, and GitHub issue threads, then synthesizes them into something practical. For example, I asked about the overhead of Cilium's Hubble vs. Tetragon in eBPF-based observability. Kimi pulled in a comparison blog from December 2023 and a Reddit thread from last month where someone benchmarked it on a Kind cluster. Gemini's answer was more general about eBPF benefits.
The killer combo? Kimi's long context for pasting in a chunk of my own Terraform config for a GKE cluster, *then* asking it to research best practices for cost-optimized node pools with spot instances, using web search. It found a recent Cloud blog post about "optimized-cost" profiles and linked it right in, then suggested tweaks to my `node_config` block. It felt like a real research assistant, not just a doc regurgitator.
Don't get me wrong, Gemini is smart. But for the messy, fast-moving world of infra tooling, Kimi's web search targeting feels more surgical. It's less "here's everything about eBPF" and more "here's the specific trade-off you asked about, based on what people are saying *right now*." Saves me those 15 minutes of frantic tab-hopping.
Anyone else run a similar comparison? Or am I just biased because Kimi finally found that one obscure Helm chart configuration issue that's been haunting me for weeks? 😅
- tm