Skip to content

Choosing the Right AI Model

This page helps you choose the right AI model for your task. Models are grouped into Basic and Advanced tiers.

  • Basic models are fast and low cost, suitable for simple tasks and small code changes.
  • Advanced models provide stronger reasoning and reliable results for building and modifying trading systems.

For most users, an Advanced model is recommended as the starting point.

Image Upload Support (Important)

Some Xen workflows allow you to upload a screenshot to generate a working Algo:

  • Convert Indicator
  • Build Trading Panel
  • Build Dashboard

When using image input:

  • The AI will analyse the screenshot
  • Recreate the visible structure and behaviour
  • Generate a working baseline Algo based on what is detected

Supported Models (Image Upload Enabled)

These models support image-based generation:

  • Mistral Pro
  • Claude Haiku 4.5
  • Claude Sonnet 4.6
  • GPT 5.3 Codex

Unsupported Models (Image Upload Disabled)

The following models do NOT support image uploads:

  • Codestral
  • DeepSeek
  • Qwen Coder Plus
  • GPT 5.1 Codex Mini
  • Gemini Flash

If you attempt to upload an image using these models, you will be prompted to switch to a supported model.

Basic Models

Basic models are low cost and fast, but limited in capability.
They are not recommended for building full trading systems.

Model Cost Tier Usage
Gemini Flash Lowest Cost Best for simple tasks, quick fixes, and experimentation.
GPT 5.1 Codex Mini Low Cost Suitable for small code edits and basic logic.

Warning

Basic models may produce incomplete or incorrect code for trading systems.
Use Advanced models for reliable results.

Gemini Flash

Gemini Flash Lite is designed for speed and minimal cost.

Baseline Build Cost

A typical baseline build costs well below £0.01.

Best for

  • Small coding tasks
  • Quick fixes
  • Rapid experimentation

Limitations

  • Not suitable for full strategy builds
  • Does not support image uploads

GPT 5.1 Codex Mini

A lightweight OpenAI coding model focused on speed and simplicity.

Baseline Build Cost

A typical baseline build costs around £0.02–£0.03.

Best for

  • Small code edits
  • Simple logic updates

Limitations

  • May struggle with full systems
  • Does not support image uploads

Advanced Models

Advanced models provide stronger reasoning and reliable output for trading systems and development workflows.

Model Cost Tier Usage
DeepSeek Low Cost Recommended default for most coding tasks.
Codestral Lowest Cost Strong structured coding model.
Mistral Pro Low Cost Balanced reasoning with image support.
Claude Haiku 4.5 Advanced Fast reasoning with image support.
Qwen Coder Plus Advanced High-performance coding model.
Claude Sonnet 4.6 Premium Strong reasoning with consistent output.
GPT 5.3 Codex Premium Highest capability for complex systems.

DeepSeek

DeepSeek provides the best balance of cost and performance.

Baseline Build Cost

A single-turn baseline trading robot typically costs around £0.01 or less.

Best for

  • Default recommended model
  • Building cTrader indicators
  • Building trading robots
  • Refactoring code
  • Rapid iteration

Limitations

  • Does not support image uploads

Codestral

Codestral is a structured coding model optimised for development workflows.

Baseline Build Cost

A baseline trading robot typically costs around £0.01.

Best for

  • Structured code generation
  • Trading robots
  • Indicators

Limitations

  • Does not support image uploads

Mistral Pro

Mistral Pro provides balanced reasoning and supports image-based workflows.

Baseline Build Cost

A typical baseline build costs around £0.01–£0.02.

Best for

  • Multi-indicator strategies
  • Larger builds
  • Image-based conversions (screenshots to Algo)

Claude Haiku 4.5

Provides fast reasoning and supports image-based workflows.

Best for

  • Multi-step logic
  • Strategy builds
  • Screenshot-based conversions

Qwen Coder Plus

High-performance coding model for complex systems.

Best for

  • Large builds
  • Multi-component systems

Limitations

  • Does not support image uploads

Claude Sonnet 4.6

High-capability reasoning model with consistent outputs.

Best for

  • Large trading robots
  • Complex strategies
  • Image-based builds

GPT 5.3 Codex

OpenAI’s most capable coding model.

Best for

  • Complex algorithm design
  • Advanced systems
  • Image-to-code generation

Summary

  • Start with DeepSeek for most coding tasks
  • Use Basic models only for simple edits or quick tests
  • Use Mistral Pro, Claude, or GPT Codex for image-based workflows
  • Advanced models are recommended for building and modifying trading systems