- Consensus, a scientific research tool used by over 8 million people, has just launched the “Scholar Agent” system – a multi-agent AI built on GPT-5 and Responses API, capable of planning, reading, and synthesizing scientific evidence much like a real researcher.
- With over 220 million indexed scientific articles, Consensus’s goal is to help scientists find, understand, and connect knowledge faster, instead of wasting time on the search phase.
- The Scholar Agent divides the research process into 4 agents:
- Planning Agent: analyzes the question and lập plans of action;
- Search Agent: retrieves documents from the database and personal libraries;
- Reading Agent: reads, summarizes, and extracts information from articles;
- Analysis Agent: synthesizes results, constructs structure, and generates the final report.
- This multi-agent structure helps reduce errors and “hallucination” phenomena by clearly dividing responsibilities, preventing one agent from controlling the entire chain of reasoning.
- Each result is accompanied by a “research context pack” – a data package containing citations, metadata, and source summaries for easy user verification. If sufficient quality evidence is lacking, the system refuses to draw a conclusion.
- Consensus switched to using Responses API instead of Chat Completions to support routing between agents, which helps reduce costs, increase stability, and improve accuracy when calling the tool.
- In internal testing, GPT-5 outperformed GPT-4.1, Sonnet 4, and Gemini 2.5 Pro in tool-calling accuracy and planning capability.
- The company chooses a direct-to-consumer (D2C) approach instead of selling through universities, targeting students, researchers, and doctors who need fast access to scientific evidence.
- The new “Medical Mode” feature supports doctors in looking up clinical evidence and has been integrated into the Mayo Clinic’s medical library.
- In just one year, Consensus has grown its revenue 8-fold and expanded to over 8 million global users. All improvements continue to revolve around the goal of “hallucination-free generative AI,” prioritizing verifiable and traceable results.
- The system has a modular design, allowing for the addition of new agents such as experiment replication, chart drawing, or statistical analysis as AI models become more powerful.
📌 The Consensus platform uses GPT-5 and Responses API to shorten research from weeks to minutes. Using GPT-5 and Responses API, Consensus is redefining the scientific research process, turning AI into a research assistant that can read, understand, and verify evidence like a human. With processing speeds hundreds of times faster but still ensuring academic reliability, this platform opens up an era of “instant research,” where science advances faster thanks to AI’s ability to synthesize knowledge at super speed.
