Uploaded June 2026 | Updated September 2026, 1 week ago
Which model reigns supreme at one-shotting an agent-native Figma clone
The prompt used was: "Create a production-grade Agent-Native app that is a pixel-perfect, fully functional Figma editor clone. Follow the conventions in this repo".
Agent-Native is the repo it was generated within, which is what it scans through to understand what an "Agent-Native" app means (basically an app with first-class agent support throughout): github.com/BuilderIO/agent-native
Harness used: Pi (no customizations, just ootb)
Models tested (and their cost to generate their app):
- DeepSeek V4 Pro - $0.13
- GPT-5.4 Mini - $0.31
- Minimax M3 - $0.42
- GLM 5.2 - $0.95
- Gemini 3.5 Flash - $1.11
- Kimi K2.7 Code - $1.25
- Composer 2.5 - est. $1.25
- Grok Build 0.1 - $1.47
- Qwen3.7 Max - $1.60
- GPT-5.5 - $1.68
- OpenRouter Fusion - $2.06
- Claude Sonnet 4.6 - $2.29
- Claude Opus 4.8 - $14.19
Which model reigns supreme at one-shotting an agent-native Figma clone
The prompt used was: "Create a production-grade Agent-Native app that is a pixel-perfect, fully functional Figma editor clone. Follow the conventions in this repo".
Agent-Native is the repo it was generated within, which is what it scans through to understand what an "Agent-Native" app means (basically an app with first-class agent support throughout): github.com/BuilderIO/agent-native
Harness used: Pi (no customizations, just ootb)
Models tested (and their cost to generate their app):
- DeepSeek V4 Pro - $0.13
- GPT-5.4 Mini - $0.31
- Minimax M3 - $0.42
- GLM 5.2 - $0.95
- Gemini 3.5 Flash - $1.11
- Kimi K2.7 Code - $1.25
- Composer 2.5 - est. $1.25
- Grok Build 0.1 - $1.47
- Qwen3.7 Max - $1.60
- GPT-5.5 - $1.68
- OpenRouter Fusion - $2.06
- Claude Sonnet 4.6 - $2.29
- Claude Opus 4.8 - $14.19










