THE PORTEX RESEARCH BLOG

Introducing COMPOSITE-STEM: an open source benchmark testing agents on frontier scientific tasks

Introducing COMPOSITE-STEM, 70 expert-curated agentic tasks across Physics, Biology, Chemistry, and Math, compatible with the Harbor Framework.

Operationalizing Expert Preferences for Model and Agent Evals

In our recent paper, we introduce AsymmetryZero, a framework for operationalizing human expert preferences as semantic evals.

Evaluating Agents on Portex with Harbor

Portex has integrated with Harbor, a popular framework to test agents on realistic workflows.

Data Valuation as a Foundation for AI Progress

Data has been called the world’s most valuable resource and “unreasonably effective” at modeling the world around us, yet, the economics governing AI data acquisition have remained largely unchanged. Up until recently, AI training data was either taken as a given (through canonical datasets like ImageNet or Common Crawl)

What Hugging Face Reveals About the Future of Model Fine-Tuning

The recent launch of OpenAI's gpt-oss series shows that performant open-weight models are abundant. Differentiation now comes from high-signal data. Hugging Face’s growth makes this visible at scale and points to a coming market for proprietary datasets for fine-tuning.

Saving 180 developer hours with RFPs

Showcasing Requests for Proposals (RFP) on the PortexAI Datalab