Ron Lancaster

Thoughts on tech and leadership

A few weeks ago, OpenAI dropped the price of o3. Because of that, it’s now cheap enough to use as my daily coder.

I use Aider with o3 in Architect mode: which directs o3 to make a plan and then 4.1 to make the edits. It’s a great combination.

The o3 model uses fewer “thinking” tokens than Gemini, which means that even though Gemini is cheaper for total tokens, o3 is less expensive to use overall:

Model Percent Correct Cost Command Correct Edit Format
gemini-2.5-pro-preview-06-05 (32k think) 83.1% $49.88 aider --model gemini/gemini-2.5-pro-preview-06-05 --thinking-tokens 32k diff-fenced
o3 (high) 81.3% $21.23 aider --model o3 --reasoning-effort high diff
gemini-2.5-pro-preview-06-05 (default) 79.1% $45.60 aider --model gemini/gemini-2.5-pro-preview-06-05 diff-fenced
o3 (high) + gpt-4.1 78.2% $17.55 aider --model o3 architect
o3 76.9% $13.75 aider --model o3 diff
gemini-2.5-pro-preview-05-06 76.9% $37.41 aider --model gemini/gemini-2.5-pro-preview-05-06 diff-fenced
gemini-2.5-pro-preview-03-25 72.9% aider --model gemini/gemini-2.5-pro-preview-03-25 diff-fenced