ChatGPT model?

[faction12-oss]

What ChatGPT model are you guys using? When I use Sol with Medium thinking, I burn through almost the entire 5-hour limit. Can anyone recommend the best setup or a better solution?

[aimanmalib]

OpenMontage does not prescribe a model. Any assistant that can read files and run code works, so what drains the limit is mostly how much composing the agent does per video, not which ChatGPT model you pick.

The repo says so itself. AGENT_GUIDE.md has two ways to build a video:

• Templated: stock scene types and components. The guide calls it "fast and reliable".
• Atelier: a hand-authored composition, which the guide defaults to for hero work like marketing and launches. It notes that atelier "costs more tokens and iteration than templated".

So two things to try with your current setup:

1. Ask for the templated path for drafts and everyday videos, for example "use the templated scene types, no atelier", and keep atelier for the one video that has to stand out.
2. Lower the reasoning effort for this project in Codex. In ~/.codex/config.toml, or .codex/config.toml inside the OpenMontage folder once you trust the project, set model_reasoning_effort = "low". The available levels depend on the model, see the Codex config reference.

If a run still eats most of the window, it would help to say which pipeline you ran and roughly how many scenes, so others can compare.

[stinson-pocket]

What ChatGPT model are you guys using? When I use Sol with Medium thinking, I burn through almost the entire 5-hour limit. Can anyone recommend the best setup or a better solution?

I have used Luna with no problems