# Replit releases benchmark testing agents as software products grow

_Published Wednesday, October 28, 2026 at 3:00 PM EDT · AI · Latest · Tier 2 — Notable_

![Replit releases benchmark testing agents as software products grow — Primary](https://cdn.sanity.io/images/bj34pdbp/migration/e3695b426a6a67d5423ff5d0e557a87c5b1789ac-3840x2160.png?w=1200&h=630&fit=max&fm=jpg)

Replit released ViBench 2.0, a benchmark that tests whether AI agents can build software products from an empty project and add up to eighteen features. It covers seventeen products, 241 feature requests and 544 test plans; eight products are public, while nine are held out.

Replit says Fireworks' runs found GPT-6.1 Sol passed every test plan in 65% of builds, followed by GPT-6 Astra at 60% and Opus 5.5 at 49%. A grading agent checks the completed products in a browser. All builders use the same OpenHands-based harness, which may disadvantage models tuned for other setups.

## Sources

- [Replit Blog](https://replit.com/blog/vibench-2-long-horizon-software-benchmark)

---
Canonical: https://techandbusiness.org/newswire/nwEC59nXGeWhOdG8HoPqO1
Published: 2026-10-28T19:00:59.484Z
Story chronology: 2026-10-28T18:50:00.000Z
Retrieved: 2026-10-28T22:09:03.051Z
Publisher: Tech & Business (techandbusiness.org)