Did a Diffusion Model Just Beat Claude Haiku?

0 Просмотры

  • Спасибо! Поделитесь с друзьями!

    Вам не понравилось видео. Спасибо за то что поделились своим мнением!

aii
Издатель
Most AI models write one token at a time, but Mercury 3 drafts many at once and claims over 1,000 tokens a second.
Inception released Mercury 3, a diffusion language model, on October 19, 2026. In Inception's own benchmark table (medium settings) it beats Claude Haiku 5.5 on 10 of 12 tests, including GPQA Diamond (89.23 vs 84.51) and IFBench (76.84 vs 54.65), and lists 1,068 tokens per second against about 240 for Haiku. Haiku still wins on DeepSWE coding (51.62 vs 46.31) and MRCR v2 long context. These are Inception's own numbers, not yet independently reproduced, and OpenRouter's live median for Mercury 3 in its first week was about 243 tokens per second.

Sources: Inception's Mercury 3 launch post and benchmark table, Inception's API docs, OpenRouter's Mercury 3 page, and CTOL Digital's live speed report.

#mercury3 #inception #claude #haiku #diffusion #ai #llm #aimodels #aibenchmarks #shorts
Категория
AI Картинки
Выполните вход или зарегистрируйтесь для отправки комментария.
Комментариев нет.