How To Use This Page

This tracker is not trying to decide which model is "best." It is trying to show where benchmark claims actually come from, which benchmarks are being foregrounded, and what kind of release surface made the claim. That matters because the same company can cite one set of benchmarks in a GitHub repo, another in a product page, and a third in a media interview.

Verified Claim Surfaces

What Makes Asian Benchmark Claims Distinctive

Three patterns stand out. First, Chinese model teams frequently publish benchmark tables directly inside GitHub repos, which makes the claim surface relatively easy to audit. Second, South Korean release pages tend to emphasize local-language strength and usable enterprise capabilities alongside general benchmarks. Third, the most informative claims are often the ones that reveal the company's real product strategy: multilingual strength, tool use, coding, agentic work, or evaluation breadth.

How To Read These Claims Correctly

• Check whether the benchmark table lives on an official repo, an official product page, or only a press interview.
• Check whether the claim is attached to a specific variant, such as instruct, thinking, or experimental mode.
• Check whether the release is emphasizing one benchmark because it flatters the model's real commercial strength.
• Use language and deployment benchmarks seriously in Asia, because they often matter more than one leaderboard headline.