Deep Research and Pro-Forma now use the newest models from each provider:
- Google Gemini 4 for the most demanding steps — the pro-forma's baseline and projections, and edits that change several figures at once. These steps can take a little longer, and their credits reflect the stronger model.
- Gemini 3.8 Flash and Gemini 3.5 Flash-Lite for everything else, including chat.
- OpenAI GPT-6 for runs set to OpenAI.
Runs set to Anthropic Claude are unchanged.