Llama 3 and 3.1 made open weight the default baseline

Llama 3 and Llama 3.1 changed the starting point for model comparisons.

Open weights used to be easy to place in a separate category: useful for certain teams, interesting for experimentation, and not necessarily part of the first production conversation. These releases made that separation harder to defend. The open route became something worth testing before choosing a hosted provider.

The point is not that open weights win every workload. They change the questions a team can ask. A real comparison now includes deployment control, latency, hardware, privacy, update responsibility, and operating cost alongside output quality.

I think of the baseline as a small test matrix:

hosted frontier
open weight self-hosted
quantized self-hosted
small local fallback

Once those routes are measured against the same examples, the decision becomes easier to explain. A hosted model may win on capability or support. A local model may win on latency, privacy, or cost. A quantized version may be good enough for a narrow workflow. The important part is that the tradeoff is observed rather than assumed.

That also makes the open route useful even when it is not selected. If it fails on the tasks that matter, the team has evidence. If it performs well, the team has to name the reason for paying for a hosted route: managed updates, operational simplicity, policy controls, or a capability gap.

Llama 3 and 3.1 did not make deployment choices ideological. They made open weights practical enough to belong in the first evaluation pass. That is the baseline shift.