Three updates this week, each adding a different kind of value to the platform.
Quarex now has a dedicated reports hub at quarex.org/reports/. This is where research, announcements, and ecosystem updates will live from now on.
The reports section is organized by category:
Every email update I send will also be archived here, so you can always find past announcements.
Visit: quarex.org/reports/
I ran a structured benchmark across seven leading AI models to see how they perform on Quarex-style questions — the kind of open-ended, contextual queries the platform is built for.
Models tested:
65 questions across 11 dimensions: accuracy, depth, nuance, structure, synthesis, source use, intellectual honesty, practical value, engagement, accessibility, and originality.
Key finding: Claude 3 Opus ranked first overall, with GPT-4o and Gemini 1.5 Pro close behind. The smaller/faster models (Sonnet, Flash, Mini) performed well on straightforward questions but struggled with synthesis and nuance.
Full methodology, raw scores, and model-by-model breakdown available in the report.
View: LLM Benchmark v1.3 - Full Results
This week's scrape of Ballotpedia brought in significant candidate changes across all three election types:
Breakdown by election type:
Notable: Ohio House saw the biggest movement with 44 new candidates entering races.
View: Candidate Changes - February 6, 2026
The weekly candidate updates are now fully automated with Python scripts that scrape, compare, and generate reports. This means faster turnaround and consistent tracking through the primaries.
The reports section will continue to grow as I publish more research. Next up: a deeper analysis of how different AI models handle politically charged questions.