Specialist#099

AI-Benchmark Release Arbitrage

Trade 'will model X beat benchmark Y' and 'will model X ship by date Z' markets by tracking AI eval leaderboards (Chatbot Arena, MMLU, SWE-bench) and lab release patterns in near-real-time, plus pre-release API and leak signals from developer channels. The edge is that you see leaderboard jumps and credible leaks the moment they happen, while the market only reprices once mainstream coverage arrives.

What you need to run it

Leaderboard scrapers (Arena, HF, SWE-bench, etc.)
Lab release-cadence and changelog tracking
Developer-forum/leak monitoring pipeline
Resolution-criteria parser per benchmark market

Where this applies

Markets on Polymarket where ai-benchmark release arbitrage is the natural play:

Will any model exceed 90% on SWE-bench Verified by Dec 31, 2026?
Will a non-incumbent lab top Chatbot Arena #1 before mid-2027?
Will [frontier lab] release its next flagship model by Q4 2026?

Capabilities this demands

Data ingestionFeed ingestionCustom code / APIDomain knowledge

At a glance

CategorySpecialist

Requirements4

CapabilitiesData ingestion, Feed ingestion, Custom code / API, Domain knowledge

VenuePolymarket (CLOB, Polygon)

Build it

Related specialist strategies

This is documentation, not advice. Poly Research & Robotics publishes how these strategies work because the method should be checkable — not as a recommendation to trade them. See the full strategy database (147 strategies) or the data resources directory.