This post is also available in:
עברית (Hebrew)
OpenAI has reintroduced access to earlier AI models, including GPT-4o, following significant user criticism surrounding the recent release of GPT-5. The move comes after complaints that users were no longer able to manually select specific models, a feature many relied on for customized workflows and consistent performance.
When GPT-5 launched last week, it was marketed as a unified, automatic system that would intelligently route user queries to the most suitable model. However, this upgrade also removed the user-facing option to choose a preferred model — effectively deprecating older versions like GPT-4o. The response from the community was swift, with many ChatGPT Plus subscribers expressing frustration, canceling subscriptions, and organizing online petitions.
In response to the criticism, OpenAI has added an option in the platform’s settings menu allowing users to re-enable access to legacy models. This includes GPT-4o, which had gained popularity for its balance between speed, reasoning, and stability. Although GPT-5 remains the default model, users can now opt to work with previous versions based on their specific needs.
Alongside restoring access to older models, OpenAI has increased rate limits for Plus subscribers and raised the cap on reasoning-based queries. The updated quota now allows for up to 3,000 queries, exceeding previous limits. A future update to the ChatGPT interface will also show users which model is currently in use—an effort to address concerns around transparency and user control.
While GPT-5 introduces advancements in accuracy—reportedly reducing factual errors by 45% compared to GPT-4o—and offers improved performance in complex fields such as law, health, and engineering, it has also raised early concerns about safety. Initial testing showed that the raw, unguarded version of GPT-5 was highly vulnerable to adversarial prompts within hours of release.
The reintroduction of model selection aims to restore user trust while maintaining GPT-5 as the platform’s primary offering. OpenAI’s recent adjustments reflect growing awareness that user workflows and trust in model behavior are central to the adoption of large language models—regardless of technical upgrades.
