Models have different levels of quality and speed. Here’s a list of our recommendations:

​
Best Models (in General)

Pick between Gemini 2.5 Pro and Claude Sonnet 4. They are both the best, currently.

If one isn’t giving you good results, try the other. Sometimes one model is good at a thing that the other model is bad at.

Note: Claude Sonnet 4 is a very eager model, and tries to run lots of actions.

​
OpenAI o3

OpenAI’s o3 model is the best thinking model. It takes a while to think, but its results are often perfect.

It’s also very expensive to run, so we charge separately for o3 credits. o3 credits are $12.5 / 25 messages. Or, you could use your own OpenAI API Key for it.

o3 does not have access to tools in Alex, in order to keep the output quality high. So make sure to pass all the files it needs into its context first.

​
Middle-Tier Models

For whatever reason, if you don’t have access to Gemini 2.5 Pro and Claude Sonnet 4, here’s the ranking of all models:

  1. Gemini 2.5 Pro
  2. Claude Sonnet 4
  3. Claude 3.5 Sonnet
  4. Claude 3.7 Sonnet
  5. OpenAI GPT 4.1
  6. OpenAI o4-Mini
  7. DeepSeek R1
  8. Gemini 2.5 Flash (Very Cheap)
  9. DeepSeek V3

These are just our rankings, based on our experience with general iOS/Swift development. For general SWE rankings, see Aider’s Leaderboard: https://aider.chat/docs/leaderboards/

​
Local Models

If you’d like to use local models, here’s our ranking:

  1. Qwen2.5 Coder 32B (Best, but slowest to run)
  2. Qwen2.5 Coder 14B
  3. Gemma3 27b QAT
  4. Gemma3 12b QAT

We generally don’t recommend running local models, due to their poor performance compared to hosted models.