HN Asks: Where Open Models Actually Fall Short
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- The poster frames a recurring industry claim that open-weight models roughly six months behind the frontier are "good enough for the majority of work" and asks the HN community for counterexamples
- The comparison targets specific pairings: GLM, DeepSeek, Kimi, and Qwen as the open-weight contenders versus Opus, Fable, and GPT as the frontier alternatives
- The response template the poster supplies asks for the task description, how the cheap model failed, whether the frontier model actually succeeded, and whether a frontier-minus-one model would have handled it in hindsight
- The poster explicitly invites reverse cases where open-weight models handled something frontier models could not, signaling an openness to disconfirming the "frontier only" narrative
Why it matters: The "good enough" thesis for cheaper open-weight models shapes enterprise build-vs-buy and cost-of-intelligence decisions; this thread aims to populate the missing real-world evidence base with named tasks and named model failures rather than vibes.
