Take this as both feedback and complaint.
The model churn within Chat is ridiculous. We don’t need the frontier, token hungry, benchmark maxxed model every three months. We need models to remain accessible, consistent, and relational.
The majority of users, per Dev Day numbers, are not codex users. They are people that rely on models for their workflow, life management, business management, creativity, and companionship.
Model churn is nothing but disruptive. It’s not like the models are the same personalities or operation wise. Prompts that worked in one model won’t return the same output in another. While that might be considered desirable, when you spend months adapting to a model’s style, quirks, and develop prompts that work well, only for the model to be retired and the models left aren’t 1 for 1 replacements, where does that leave you?
Why are users expected to waste their time and energy because the company that had models available for nearly two years (4o) and a year (4.1) decided that we need something shiny and new every three months.
The API is not the answer for the vast majority of users. There is no simple porting of our workspace/chats/memories/context/etc from the chatgpt environment to a third party harness.
OAI hasn’t got an API harness that we can use as a base that provides the same functionality as the app/web, which would make moving to API much easier.
FEEDBACK/SUGGESTION:
- OAI creates a harness that allows users to move their entire account content to API.
Make the port seamless enough that we can continue using preferred models. - Allow API access within ChatGPT.
Let users pay for API credits, or use their Codex/Work credits to bring the API model of their choice back into the ChatGPT environment to continue working with it. - Create an Enduring Model Access Add-on for all plans.
Let users opt-in to paying a little more to continue having access to the model as it was within the app. They pay this on top of their current plan. The model is served as it was, allowing users to continue work that relied on a specific model.
This doesn’t impact update of newer models, as users that cannot work with those models either stop using the service, or move to a different provider anyway.
People that want to use the newest models will continue to do so, as the options are all user choice. The user has to actively CHOOSE to opt-in.
These models are still being offered via API anyway, why not allow them to stay active where users actually want them?