One-Shot Build
A benchmarking paradigm requiring an AI model to generate a complete, complex output (e.g., full product documentation) from a single, comprehensive input without iterative refinement or follow-up queries.
Key Characteristics
- Tests model’s ability to synthesize large, multi-faceted inputs in one pass
- Emulates real-world scenarios where prompt engineering is impractical
- Measures holistic understanding beyond simple task completion
Related Comparisons
- Claude Opus 4.5 and ChatGPT 5.2 were evaluated on a “One-Shot Build” benchmark using a massive Product Requirements Document (PRD) for the Showbiz app
- Input: Comprehensive documentation folder containing technical specs, design tokens, and personality guidelines
- Task: Generate a functional PRD from raw input without iterative feedback
- Uses an “Impossible” PRD designed by Matt Maher to push model boundaries
Backlinks
- 2026 04 14 Compare of Claude Opus 45 vs ChatGPT 52 Matt Maher
Source Notes
- 2026-04-23: https://www.youtube.com/watch?v=iUzrE3-FHgA Summary of comparison between OpenAI’s GPT-5.2 models and Anthropic’s Claude Opus 4.5 using a complex “One-Shot Build” benchmark.