LongCat first impression
LongCat came out with no standout benchmarks, no hype, nothing special on paper.... and made by Meituan, a food delivery company.
I've stopped judging models by benchmarks. So I gave it a real task in OpenChamber: build a cute agent UI with smooth UX and real polish.
What surprised me wasn't that it built something decent. It's how it built. LongCat doesn't assume. It writes clearly, concisely, and checks in with you before making decisions. It does one step at a time and gives you space to provide feedback and steer before it continues. It doesn't overbuild. it asks. In the world of "one shot tests" this is bad. In real agentic engineering, this is very very good.
This model provides a genuinely nice experience. It's not the cheapest (Meituan sells token packs, and you're paying a premium over DeepSeek or GLM 5.3 Flash), but the interaction quality is there.