TLDR Tech / AI
Can AI agents build real Stripe integrations? We built a benchmark to find out
Stripe released a new benchmark to test whether AI agents can autonomously build complete payment integrations, finding that Claude Opus 4.5 achieved a 92% success rate on full-stack tasks while OpenAI's GPT-5.2 scored 73% on backend problem sets. The 11-task benchmark revealed that while leading models could navigate UIs and debug live issues across an average of 63 turns, they still struggled with ambiguous situations and occasionally got stuck during browser-based workflows, highlighting the gap between code generation and production-ready software engineering.
Read the full story