Claude Haiku 5.5 reportedly reaches 91% of GPT-6.1 Sol’s design-task score at 5% of its estimated cost

OpenDesignHQ says its benchmark used everyday design tasks from real user requests and found Haiku 5.5 reached 91% of GPT-6.1 Sol’s score at 5% of its estimated cost. It also says Haiku cost less than DeepSeek V4.1 Flash and finished fastest. Separately, a user claims scanning one smart contract for vulnerabilities costs about $1.22 and estimates that rescanning every contract in a protocol weekly costs less than one human audit hour.