EvoWeb benchmark: 320 sequential requests expose coding agents' cross-request integration failures
TD-Labs · hf · 2026-10-20
EvoWeb is a new benchmark for coding agents on evolving web applications: 320 cumulative requests over 32 web apps grounded in real product features, structured by functional dependency graphs. Agents iteratively modify a persistent codebase while a multimodal web agent tests each intermediate app in a live browser, scoring new functionality, cross-request integration, and regressions. Experiments across frontier open- and closed-source models show clear performance gaps, with cross-request integration as the key challenge; deeper-dependency requests are solved less accurately and runtime errors grow over successive requests.
More from coding & agent
- Editor says Grok handled 90% of a music video edit — shares her 3 workflow tips — chaitu · 2026-10-20
- OpenAI Devs teases new Codex showcase: cloud environments and terminal chat — OpenAIDevs · 2026-10-20
- Cua launches Spaces SDK to run Claude Code and any agent on compute you choose — Scobleizer · 2026-10-20
- Dev: top AI agents do more backup and verification on DB migrations than 99.99% of engineers — tristanbob · 2026-10-20
- Build a Weekly Report Agent: Day 1 of an Agent-a-Day Series for Scientists — nlarusstone · 2026-10-20
- 20 questions to answer before letting agents touch production data — Then_Respect_1964 · 2026-10-20