When a new model is released, it is interesting to test it out to see how the LLM landscape is changing. If you are considering buying Anthropic IPO stock (I am not) this can also be worthwhile. I tried Kimi K3.1 Pro on OpenRouter and had it analyze a small Rust function.
Kimi K3.1 Pro took about 4 minutes to analyze the code and returned a simpler version. It was not sure if the new Rust function would be faster, due to the complexities of compiler optimizations. The cost was about $0.08, which is somewhat expensive for this sort of task.
• The performance gain from the Rust function (when it can be measured) is tiny (0.1% in a micro-benchmark).
• The value of Kimi K3.1 Pro is not in its performance online, or its costs, but that it will be available as an open-weights model available to run locally.
• As far as Anthropic goes, I would be concerned—open-weight models like Kimi K3.1 seem to have comparable results.