Every Enterprise Should Leverage Opensource Models
Qwen Code 3.8 Max is a better value than Claude Code, but only on par/close to par with Claude Code from a performance standpoint.
Neither Qwen, nor Kimi, nor GLM have surpassed Claude Code from a performance standpoint. Nobody is ahead of Anthropic in terms of pure performance. Yet enterprises are diversifying away from Anthropic and OpenAI for a variety of reasons - cost, wanting to control model weights, wanting to control their own IP (not exposing to Claude or OpenAI's caching).
There is no reason for an Enterprise to use Anthropic or OpenAI in my view - every Enterprise should want to control the model weights and lock-down their own IP.
Yet, I’m not sure how many companies are even aware of the IP issue. Anthropic’s largest customer in the early years was Cursor, Anthropic saw Cursor was working, saw Cursor’s underlying code, and built Claude Code. Anthropic partnered with Accenture, then rolled out an IT Services business (earlier this week IBM announced a partnership with Anthropic, “Hello, is anyone paying attention?”).
The IP theft is even more explicit now given that Anthropic’s Fable 5 holds all customer prompts for 30 days. I assume OpenAI is doing something similar, but haven’t investigated as I do not have access to OAI’s bleeding edge model.
Here’s my recent article - caching and stealing.