Model Selection Guide
This guide explains when to use each available AI model in CodePilot. Choosing the correct model improves accuracy, cost efficiency, and output quality. We provide both the most advanced and budget friendly models.
Latest AI Models as of January 2026
Model Selection Guidance
Models in the selection dropdown are ordered by cost, with the cheapest lightweight models at the top and the most expensive advanced models at the bottom.
It is best to start with a lower cost model like DeepSeek Coder and only switch to a higher tier model if you need improved reasoning or more complex output.
You can change the model at any time during a conversation.
GPT-5.2
GPT-5.2 is OpenAI's flagship model family designed specifically for professional knowledge work, enterprise applications, and long-running, complex, agentic workflows, which is great for coding cTrader indicators, cBots and Plugins.
Advanced Model
Best for
- Building new advanced trading tools from scratch.
- Algorithm Strategy design and rule definition.
- Conversions where logic must be interpreted, not translated.
- Multi step workflows and safety checks.
Why
- Strong reasoning and requirement clarification.
- Good at enforcing platform rules and lifecycle correctness.
GPT-5 Mini
GPT-5 mini is a lightweight, cost-efficient, and high-throughput variant of OpenAI’s GPT-5 flagship model. Cost-optimized reasoning and chat; balances speed, cost, and capability
Lightweight Model
Best for
- Basic trading tools.
- Simple modifications.
- Small feature additions.
- Minor logic tweaks.
- Cost efficient experimentation.
Why
- Faster and cheaper
- Adequate when logic is already well defined.
- Not suitable for complex reasoning or architecture decisions.
GPT-5.2 vs GPT-5.2 Mini
GPT-5.2 is designed for tasks that require deeper reasoning, structured thinking, and multi step decision making, such as building new indicators or cBots, interpreting trading logic, performing complex conversions, or enforcing platform and lifecycle rules. GPT-5.2 Mini is optimised for speed and cost efficiency and is better suited to smaller, well defined tasks such as minor code fixes, parameter changes, or simple logic adjustments where the overall structure is already correct and no architectural reasoning is required.
GPT-5.2 vs GPT-5.2 Codex
GPT-5.2 is a general reasoning model designed for understanding intent, clarifying requirements, and handling multi step logic, making it suitable for new builds, strategy design, complex conversions, and workflows where correctness and structure matter. GPT-5.2 Codex is a code focused model optimised for reading, modifying, and refactoring existing source files and is more appropriate when requirements are already fixed and the task is to apply precise, minimal changes to C# code without reinterpreting the design.
Claude Haiku 4.5
Lightweight/Advanced Model
Claude Haiku 4.5 is a fast, cost-effective, and highly intelligent "small" model from Anthropic, delivering near-frontier performance for coding trading tools for the financial markets.
Best for
- building basic to advanced trading tools from scratch.
- Users who want concise, readable output.
Why
- Very fast response times.
- Clear explanations.
- Less strict about platform guardrails.
Limitations
- Can miss subtle API or lifecycle constraints.
- Not ideal for complex trading logic.
Claude Sonnet 4.5
Claude Sonnet 4.5 is one of the strongest coding models available, offering excellent reasoning, code quality, and consistency, which makes it well suited for developing complex trading tools and algorithms across modern trading platforms.
Advanced Model
Best for
- Building new advanced trading tools from scratch.
- Algorithm Strategy design and rule definition.
- Conversions where logic must be interpreted, not translated.
- Multi step workflows and safety checks.
Why Sonnet 4.5
- Strong comprehension and structured output.
- Good middle ground between speed and depth.
DeepSeek V3.2
DeepSeek Chat V3.2 is DeepSeek's latest flagship model specifically designed for professional knowledge work, technical applications, and complex reasoning workflows, making it excellent for coding cTrader indicators, cBots, and Plugins, as well as various other technical and analytical tasks.
Advanced & Cheapest Model
Best for
- Building new advanced trading tools from scratch.
- Algorithm Strategy design and rule definition.
- Code Generation and Analysis - Strong capabilities in multiple programming languages
This model is particularly well-suited for developers, analysts, and professionals who need sophisticated reasoning capabilities without the cost barriers of some premium models.
Claude Haiku 4.5 vs Claude Sonnet 4.5
Claude Haiku 4.5 is optimised for speed and concise output and is best suited to quick explanations, small fixes, and simple conversions where requirements are clear and the logic is straightforward. Claude Sonnet 4.5 provides deeper reasoning and stronger code comprehension and is better suited to medium complexity feature additions, cross language conversions, code review, and refactoring tasks where structure, consistency, and accuracy matter more than raw response speed.
Workflow Mapping
Experiment with DeepSeek Coder v3.2, we found it excellent for coding tasks at x10 cheaper than other models.
| Workflow | Recommended Models |
|---|---|
| Convert Indicator | GPT-5.2, Claude Sonnet 4.5 |
| Build Indicator | GPT-5.2 |
| Build cBot | GPT-5.2 |
| Fix Existing Code | GPT-5.2 Codex, GPT-5.2 Mini, Claude Haiku 4.5, Claude Sonnet 4.5 |
| Add Alerts | GPT-5.2 Mini, Claude Haiku 4.5 |
| Add Risk Management | GPT-5.2, Claude Sonnet 4.5 |
| Build Plugin | GPT-5.2 |
| Free Typing Mode | Any model, advanced users only |
Notes
- Use lighter models when requirements are clear and changes are small.
- Use larger reasoning models when logic, structure, or safety rules matter.
- CodePilot provides guidance but does not lock model selection.