Choosing an LLM Provider

Best Practices → Choosing an LLM Provider

---

Provider/Model Breakdown

Model	Strengths	Best For
Anthropic Claude 3.5 Sonnet	- Exceptional use of UE C++ patterns - Macro usage - Great documentation	Complex systems, Performant code
Anthropic Claude 3.5 Haiku	- Use of UE C++ patterns - Macro usage - Minimal or no documentation - Fast	Less complex systems, Prototypes, Affordability
OpenAI GPT o3-mini	- Great use of UE C++ patterns - Macro usage - Great documentation	Complex systems requiring thoughtful logic reasoning, Performance-minded code, Great pricing
OpenAI GPT o1	- Exceptional use of UE C++ patterns - Macro usage - Exceptional documentation	Complex systems requiring advanced logic reasoning, Performance-minded code
OpenAI GPT-4o	- Decent use of UE C++ patterns - Moderate code structure	Medium complexity translations
OpenAI GPT-4o-mini	- Inconsistent use of UE C++ patterns - General code structure - Fast	Less complex translations, Prototypes, Affordability
OpenAI GPT o3-mini	- Great use of UE C++ patterns - Macro usage - Great documentation	Complex systems requiring thoughtful logic reasoning, Performance-minded code, Great pricing
Gemini 2.0 Flash Thinking	- Great use of UE C++ patterns - Macro usage - Decent documentation	Complex systems requiring advanced logic reasoning, Performance-minded code, Massive 2M token context window
Gemini 2.0 Pro	- Good use of UE C++ patterns - Macro usage - Good documentation	Complex systems, Performance-minded code, Massive 2M token context window
Gemini 2.0 Flash	- Decent use of UE C++ patterns - Macro usage - Good documentation	Fast, Complex systems, Performant code, Huge 1M token context window
Gemini 1.5 Pro	- Adequate use of UE C++ patterns - Okay documentation	Less complex systems, Massive 2M token context window
Gemini 1.5 Flash	- Use of UE C++ patterns Okay documentation	Very fast, Quick Conversions, Huge 1M token context window
DeepSeek R1	- Adequate use of UE C++ patterns - Macro usage - Acceptable documentation	Complex systems requiring thoughtful logic reasoning, Performance-minded code, Cost-Effective Capabilities
DeepSeek V3	- Adequate use of UE C++ patterns - Moderate code structure - Moderate Macro usage - Documentation	Quick conversions, Prototypes, Affordability
Ollama	- Local processing - Wide range of model capabilities - Ability to use custom models	Full control of model, 100% Privacy, Offline work, Cost-effectiveness

Note

Claude 3.5 Sonnet and o3-mini typically produce the most idiomatic Unreal Engine C++ code with proper macro usage and memory management patterns for reasonable pricing.

Important

DeepSeek very likely does use API data for training their models, so use at your own discretion.

Model Selection Guidelines

1. Consider Your Needs

   • Translation complexity
   • Budget constraints
   • Privacy requirements
   • Quality expectations

2. Balancing Factors

   • Cost per token
   • Translation quality
   • Processing speed
   • Context window size

Note

More capable models typically cost more but require less manual refinement of the generated code.