Technology

Design

SOTA model - state of the art model; the most intelligent, advanced, up to date model available in the market (AI companies releases models regularly so you need to keep track)
- OpenAI GPT o3
- OpenAI GPT 4.1 (4.5 is available but crazy expensive and only as research preview)
- Claude Sonnet 3.7
- Google Gemini 2.5 (Flash and Pro)
- xAI Grok 3
Open-source models - not for profit, available to download and run on your computer (needs powerful machine). They are usually 1 generation behind SOTA.
- DeepSeek R1
- Meta Llama 4
- Alibaba Qwen 3
Multimodal - can take images as input
Thinking model - able to, in multiple times, self-reflect and check response before responding to user. This reduces hallucination and increases accuracy.
- DeepSeek R1
- Google Gemini 2.5 Pro (Flash is non-thinking)
- OpenAI GPT o3
- Claude Sonnet 3.7 has a "Thinking" mode
Function-calling/Tool-calling/MCP - it's just another term for giving an AI model access to tools to expand it's capabilities and use (imagine Homo erectus with fire and other tools)

Take note that multi-modal SOTA models can take a mockup (low or high, doesn't matter) and output code.