AI news ·
GPT-6 Astra cuts Parallel's research time and cost in half
Parallel cut AI agent research time and cost by roughly 50% using GPT-6 Astra, with the model delivering the same quality through fewer searches and tokens.
Parallel, a company building developer infrastructure for AI agents that perform knowledge work over the web, has cut research time and cost by roughly half using GPT-6 Astra. The model delivered the same quality of results as previous frontier models while issuing fewer research calls and consuming fewer tokens, according to tests run by the company.
Compiling complex labor data in half the time
In a test designed to stress the system, Parallel asked an agent to research six different labor-market statistics across four states over a six-month period. The agent had to search multiple websites, gather the information, and synthesize it into a single research report. GPT-6 Astra completed the work in half the time of prior models, with a roughly 50% reduction in code cost.
"With Astra, we've demonstrated that you can get the same high-quality research much, much faster with fewer research calls and less tokens," said Devin Gupta, Member of Technical Staff at Parallel Web Systems.
Fewer steps to a useful answer
Beyond raw speed, GPT-6 Astra took more focused search paths. Instead of running broad, sequential queries, the model issued targeted searches and incorporated its world knowledge to stay on task. "Astra issued more targeted search queries and focused on the ultimate task better, incorporating its world knowledge compared to previous models," Gupta said.
That tighter focus changes the economics of multi-agent workflows. Parallel found it more practical to divide research among sub-agents, with GPT-6 Astra delegating specific tasks so work happens simultaneously rather than crawling through a single long sequence of searches. For teams building AI agents that do heavy web research, this shift from sequential to parallel execution removes a major bottleneck.
Why this matters for research and operations professionals
For analysts, HR researchers, and operations managers who rely on labor-market data, the results point toward faster turnaround on multi-source research reports without sacrificing accuracy. A task that once tied up an agent - and a budget - for twice as long can now complete in roughly half the time at half the cost. That creates room to run more queries, compare more geographies, or respond to ad-hoc requests without blowing out the research budget. As models like GPT-6 Astra make AI Agent Courses and tooling more efficient, the practical ceiling on what a small research team can produce in a day rises considerably.