Price doesn't buy quality

3.3×
what Kimi K3 costs next to Claude Sonnet 5.5, for a lower score

On the public-sector benchmark, the most expensive of the leading models is not the best one. Kimi K3 costs $0.12 for each task and got 69% of the work right. Claude Sonnet 5.5 costs 3.5¢, less than a third, and got 70%.

Further down the price list, DeepSeek V4.1 Flash got 64% right for 0.43¢ a task: 8.3× cheaper than the top model, for 5 points less.

A price here is what the model charged for the words it read and wrote on our tasks, at list price. It leaves out the person who still has to check every draft, which usually costs far more than the model does.

See every model by price and score →