BenchmarkWMT25
WMT25
The evaluation task held in 2025 by WMT, the international machine translation evaluation conference. It is one of the authoritative benchmarks in the machine translation field.
Timeline (last 90 days)
Oct 21
Offline evaluation by the research team on the WMT25 benchmark showed that pairwise-judging circular voting yields a very clear improvement over pointwise best-of-N selection.
Unverified50%