Swyx casually asked: Should we have a "Long-Lasting Edition" for ChatGPT?

Title

Swyx Asks Former OpenAI Developer Lead: Should ChatGPT Have a “Evergreen” Option?

Summary

Shawn Wang (Swyx), who operates Latent Space and the AI Engineer Summit, discussed the idea of an “Evergreen version of ChatGPT” with Edwin Arbus (former OpenAI, now Cursor). The official definition of this concept has never been established; what Swyx envisions is likely a replacement of rigid version switches with rolling updates, so that models aren’t abandoned every few months. This notion typically stems from developer fatigue—every time a new model is released, they have to retest and revise prompts.

In short: this is just a casual tweet, not a product plan or a formal proposal, but it truly resonates with developers.

Analysis

Swyx has worked at AWS, Netlify, Temporal, and Airbyte, and has founded Smol AI; he has seen “version deprecation” far too many times. Arbus previously worked in the developer community at OpenAI and now creates a similar experience at Cursor. Neither elaborated on how “Evergreen” could be technically implemented.

The concept of “Evergreen” is borrowed from the software world: both Chrome and Firefox offer continuous automatic updates without major version numbers. Applied to large models, it could be imagined as an API endpoint that gradually improves but strives to avoid breaking existing integrations.

• The reality is this: developers have built things based on GPT-4, only to find GPT-4-turbo and GPT-4o arriving with different behaviors, forcing teams to repeatedly revert to testing, rewrite prompts, and reassess use cases.
• The attractive aspect of this idea: if there were an endpoint that is “always the latest and has stable behavior,” the integration costs would decrease significantly.
• But whether this can be achieved technically is still uncertain. Under current large model training and inference methods, it’s unclear if “continuous upgrades” and “maintaining compatibility” can coexist.

Impact Assessment

• Importance: Not high. This is just a tweet, not a product release; currently, there are no signs of further discussion or indication that OpenAI intends to adopt it. However, both speakers have some influence, so this topic may come up again in the future.
• Category: Developer Tools, Model Version Management

Conclusion: For this topic, we are still in the early observation stage. The ones truly relevant are the entrepreneurs and platform providers building developer infrastructure and compatibility layers, which have little relation to traders and passive holders.