Parallel says GPT-6 Astra cut research time and cost by half

Parallel said GPT6 Astra helped its agents complete webbased research tasks in less time and at lower cost than earlier models. The startup, which builds infrastructure for AI agents, used the model on a research task involving labormarket statistics across four states over six months. According to the company, the work finished in about half the time and with roughly 50% lower code cost while maintaining the same quality. Parallel also said the model made more targeted searches and reached answers in fewer steps. The company said this could make it easier to split research across multiple agents and handle more complex tasks at scale.