Enterprises want to give our employees the best value models for the time, cost, and quality required for the task they are going to use them for.
Below is a graph from Artificial Analysis of some of the most commonly used models and their value for their price as of July 2026. The black line signifies the best models for that price at that intelligence tier.
Most enterprises have contracts with the major labs like OpenAI and Anthropic who are providing the models in the green circle:
These models are some of the best models for their intelligence at the higher end of the spectrum. However, what if companies have tasks that don’t require that level of intelligence? They are stuck with paying for models that have more than enough intelligence needed for the task at hand, when instead, there exists models for these easier tasks that provide much better value for their cost. Those are the models in the red box, and notice how most of them are open source.
By acquiring access to these cheaper models that are still very capable, companies can route easy requests to these and save 10x or more on inference costs vs paying for the models from the leading AI labs.
Here’s another benchmark:
This is called Design Arena. It allows users to give design tasks like making websites to 2 random models and they blind pick which model they think they did better.
Here it is for Non-Agentic Web Dev:
And for Agentic Web dev:
Notice again, the best value models are open source ones right now.
Now, the top closed source models do tend to have better speed for output quality (this is because the open source models tend to spend a lot of time thinking), so if your business cares about speed, this could weigh your decisions:
Overall though, most basic apps and data transformation tasks that people are doing in businesses will not require speed and the extra cost to build, which makes open source models still a better value for enterprises using LLMs at scale.