Replit's ViBench 2.0: Best Models Pass Only 49-65% of Build Tasks

Replit released ViBench 2.0, a benchmark for long-horizon software building where agents construct complete products feature by feature without seeing future requirements. Even the top model, GPT-6.1 Sol, passed only 49-65% of the 17 product build tasks.

2026-10-29 ~ 2026-10-29 · 2 related posts

• Replit's ViBench 2.0: best models pass only 49–65% of builds, GPT-6.1 Sol leads — amasad · 2026-10-29
• Replit's ViBench 2.0: top models pass only 49-65% of long-horizon builds — amasad · 2026-10-29