Replit's ViBench 2.0: top models pass only 49-65% of long-horizon builds
amasad · x · 2026-10-29
Replit released ViBench 2.0, a benchmark for long-horizon software building where agents start from an empty project and build a whole product feature-by-feature without seeing what comes next. Even the best models (GPT-6.1 Sol, GPT-6 Astra, Opus 5.5) pass every check in only 49-65% of builds, with GPT-6.1 Sol in the lead. The benchmark is now part of Fireworks AI's Specialized Intelligence Index with official results from Fireworks' runs.
Related event: Replit's ViBench 2.0: Best Models Pass Only 49-65% of Build Tasks(2 posts)→
More from coding & agent
- LangChain engineer: most agent rules made it worse — deleted them after Opus 5.5 — LangChain · 2026-10-29
- His most-starred repo is now fully vibecoded and vibe-PR'd — LFAN passed in 1.5 GPU-hours — ChongZzZhang · 2026-10-29
- Vercel Launches Agentic CI With Turborepo Task Skipping, 2-6x Faster Builds — shuding · 2026-10-29
- Celesto AI lets you run GitHub Actions on its cloud computers with per-job pricing — aniketmaurya · 2026-10-29
- GitHub sees 7.4B monthly commits from AI agents, adds 4M CPU cores to cope — ai · 2026-10-29
- 5 practical tips to cut CI time when coding with AI agents — aniketmaurya · 2026-10-29