Parallel Claims to Double Web Search Efficiency with GPT-6 Astra

Parallel reports having halved the time and cost of its web research thanks to GPT-6 Astra, while maintaining quality. The company highlights more targeted queries, fewer steps, and the ability to delegate to sub-agents to work in parallel.

Parallel emphasizes delegation to sub-agents and reduced wait times

Parallel states that increased efficiency makes it more practical to distribute a research task among multiple agents. The company claims that GPT-6 Astra can delegate specific tasks to sub-agents, allowing for simultaneous work and reducing the time spent on a single sequence of searches. It reports now having a better pathway from a complex question to a sought-after answer, with less waiting, reduced costs, and more opportunities for demanding large-scale work. Parallel also notes that it has observed more targeted searches requiring fewer steps to achieve a useful result.

A six-month test of statistics across four states

In an experiment, an agent was tasked with collecting six distinct labor market indicators across four states over a six-month period. The process involved visiting multiple platforms, extracting data, and then compiling everything into a single report. According to Parallel, GPT-6 Astra completed this task in half the time of previous models, with approximately a 50% reduction in coding costs, while maintaining the same quality. Devin Gupta adds that Astra allows for achieving the same quality more quickly, with fewer search calls and fewer tokens.

Quantifiable gains and technical context at Parallel

Parallel indicates it has halved the time and cost of research compared to its older models, specifying that labor market analyses were conducted twice as fast and for half the price. The company observes a significant improvement in terms of speed and costs thanks to GPT-6 Astra, whereas the most time-consuming tasks previously required a larger model and in-depth reasoning, which increased resource consumption. Parallel is developing an infrastructure for agents operating on the web, with tools covering web grounding for voice agents and research for financial and legal clients, combining cutting-edge models and web navigation. Devin Gupta emphasizes that Astra issues more targeted queries and focuses better on the ultimate task by integrating its world knowledge than previous models.