Reddit user reveals which AI models survive a month of real work

In a viral Reddit post, user derspenti calls out the uselessness of day-one impressions and instead reports on which AI models are still in his stack after a month of real work. Two models survived: Qwen3.6 27B for any task requiring factual knowledge, and Ling-3.0-flash, which he initially placed in an agent executor slot expecting it to fail but it never did. He notes Ling is now free on OpenRouter and confirmed open source soon. Two other models he was initially excited about were quietly removed, though he declines to name them to avoid arguments.

Derspenti’s key insight is that the highest praise is a model you stop thinking about—one that just works for a specific task without needing attention. He also invites stories of models that got worse over time or were kept out of inertia. The thread highlights a growing demand for real-world longevity over launch-day hype, especially among professionals building agentic workflows where reliability and consistent performance matter more than flashy benchmarks.