Notices
Not just new features. This also covers values we got wrong and fixed, and policy changes. Implementation details are left out.
Added a performance column to the price comparison
Price alone is often the wrong basis for a choice, so the intelligence index now sits in the same table.
We measure Korean-language performance ourselves
Every other subject uses Artificial Analysis' published figures as-is. Korean is the one we measure directly, and coverage is still partial — we show "not tested" separately from a low score.
Fixed non-reasoning models inheriting reasoning-model scores
Some models were showing scores from a sibling model's test they never actually sat. Scorecards now show only tests a model actually took, and we clarified on the page what the overall index means.
Fixed rank changes not showing for a full day
The way observations accumulated meant the rank-change arrow could go a whole day without appearing even once. It now shows precisely when the underlying score actually crosses a boundary.
Added compare and cost tools to the MCP server
compare_models lines up candidate models in one call instead of N, and estimate_cost turns usage into a monthly bill.
Token prices now show as a monthly bill
"$2.50 per million tokens" doesn't tell you much on its own. We translate it into light, steady and heavy usage tiers, in USD and KRW.
Moved the runtime to Node + PostgreSQL
Started on Cloudflare Workers, but the history the collectors need to keep and the queries over it fit a relational database better. Nothing changes on screen.
AI Wave launched
New releases, price changes, API changes and deprecations, merged from vendor announcements, GitHub, Hugging Face and the OpenRouter catalog into one feed. Live data collection and ads went live together.