Peter Yong Zhong

Researcher @ Replit, CS PhD @ CMU (Leave of Absence) || Northwestern ‘22 || 🇦🇺✈️🇺🇸

[Peter Yong Zhong @Lambda_freak]

@giansegato More to come!! One thing that we worked hard on with Vibench is that the methodology is extendable. Vibench 2 is a great step up in rigour.

[Peter Yong Zhong @Lambda_freak]

@RudrajMehta @amasad It's not just a one-shot prompt, but rather a series of feature building. Internally we look more than just the correctness rate (albeit it's the most important number), but rather a lot of the quality metrics you allude to.

[Amjad Masad @amasad]

ViBench 2.0 (prompt to app benchmark) just dropped and GPT 6.1 Sol takes the lead.

[Peter Yong Zhong @Lambda_freak]

@dusangran @amasad The beauty of ViBench 2.0 is our emulation of this type of iteration. It's not just a single PRD to a fully completed app (ViBench 1.0), but rather feature by feature build where assumptions of the software keep changing.

[Peter Yong Zhong @Lambda_freak]

@noelkimx @Replit Exactly what we want to capture in this benchmark!

[Peter Yong Zhong @Lambda_freak]

Excited to launch ViBench 2! 🚀 As builders get more ambitious, our greenfield benchmark should too. The team worked incredibly hard to keep the signal strong.

Shout out to @dryanfurman, Preeya Kirani, @Georgian_io & @FireworksAI_HQ

[Replit ⠕ @Replit]

Today we're releasing ViBench 2.0, a benchmark for long-horizon software building. Agents start from an empty project and build a whole product, one feature at a time, 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.

ViBench 2.0 is now part of @FireworksAI_HQ’s Specialized Intelligence Index, with official results from Fireworks’ runs.

Read more: https://t.co/WDbfapcWhC

[Replit ⠕ @Replit]

Today we're releasing ViBench 2.0, a benchmark for long-horizon software building. Agents start from an empty project and build a whole product, one feature at a time, 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.

ViBench 2.0 is now part of @FireworksAI_HQ’s Specialized Intelligence Index, with official results from Fireworks’ runs.

Read more: https://t.co/WDbfapcWhC

[Peter Yong Zhong @Lambda_freak]

@_jzhao Amazing insights!!!

[Peter Yong Zhong @Lambda_freak]

@ileen_fan It’s pretty good lmao

[Peter Yong Zhong @Lambda_freak]

@dryanfurman @zhenthebuilder Looking forward to cooking more with you!!

[Replit ⠕ @Replit]

Reimagining technology as a catalyst for human connection, @tigrisli designs installations, electronic hardware, and experiences that evoke curiosity and play.

See her perfume installation, transforming memories into bespoke fragrance with AI in real time, on day two of Vibecon.

NYC, June 17–18.

Get your tickets at https://t.co/X7VGQdilOX

[Peter Yong Zhong @Lambda_freak]

@amasad @samuel_spitz

[Peter Yong Zhong @Lambda_freak]

https://t.co/72jPJ9nWHE

[Peter Yong Zhong @Lambda_freak]

Much of vibe coding system design is about leveraging existing software engineering concepts and making them accessible with AI. At ACM CAIS, I presented a poster on how Replit's Agent 4 did exactly this: bringing PR/CI/merge workflows to unlock teamwork and parallelism.

[Peter Yong Zhong @Lambda_freak]

@ImenKedir @pirroh 🚀

[Michele Catasta @pirroh]

Hundreds of thousands of you unleashed your creativity on Replit Agent this weekend -- completely free.

Our infrastructure handled ~4× the usual load, with tens of thousands of agents running in parallel.

A real test of our scale, and a powerful reminder of how passionate this community is.

Thank you for building with us 🧡

[Peter Yong Zhong @Lambda_freak]

@pirroh Was super exciting to see so many users building great things! Was also a great opportunity to harden our agent infrastructure to prepare for the next 10x of growth.

[ACM Conference on AI and Agentic Systems @CAISconf]

🎤 Keynote announcement:

@trq212 (Thariq Shihipar), Member of Technical Staff on Claude Code at @AnthropicAI, is keynoting #CAIS2026.

Thariq's "Lessons from Building Claude Code" series on Skills, prompt caching, tool design, and "unhobbling" is required reading for anyone building agentic systems.

We're thrilled to have him.

📍 San Jose · May 26–29
🔗 https://t.co/BygtfeWpXM

[Peter Yong Zhong @Lambda_freak]

@heathercmiller @CAISconf Really cool to see a lot of industrial research being showcased without compromising peer review scrutiny.

[Peter Yong Zhong @Lambda_freak]

Gonna miss the hours spent setting up random creds, accounts, ends just to run the app(not even mention fighting the deployment process)

[jordwalke @jordwalke]

Introducing real iOS and Android device testing, right from your browser.

If you've ever:
👉built a @Replit mobile app and had no way to test it on a phone
👉 been solely on Windows/Android and don't touch iOS

…this is for you!