Are we wasting time building enterprise agents on open-source models? (My experience with Ling 1T 2.6)

Hey everyone, I build custom agents for enterprise clients, and lately, I’ve been questioning my entire tech stack. Recently, I spent some time testing the new Ant Ling 1T 2.6 model. Don't get me wrong—they are absolutely on the right track technically. It’s cheap, incredibly fast, and prioritizes execution. For building slick internal dashboards, handling basic coding tasks, and general speed, it’s actually pretty solid. But here’s the catch: it’s not a reasoning model. To make it work reliably in an enterprise setting, you have to aggressively optimize system prompts and heavily sanitize user inputs. You need a crystal-clear understanding of its capability boundaries, or it just falls apart. This got me thinking... is it really worth investing so much time and energy into secondary development and evaluation of open-source models? The economic upside of open-source is huge for enterprise clients, but the research and testing overhead is exhausting. Their capabilities are rarely comprehensive out-of-the-box. You have to spend days just finding the right harness. In my testing, Openclaw was pretty disappointing, though Hermes turned out to be much more stable. Because these aren't always the absolute SOTA models, you have to dig deep to find exactly what they can do and where they break. It drains so much energy just benchmarking and tweaking before you even start building the actual product. I see models like Ling and Kimi making real efforts to catch up, which is great. But I’m genuinely worried: if we pour all our resources into wrestling with open-source models to make them enterprise-ready, are we on the right path? Or are we just burning time we should be spending on actual product features? Would love to hear from other agent devs. Are you guys sticking to proprietary APIs, or is the open-source grind actually paying off for you?