OpenAI revises GPT-5 rollout after ChatGPT performance concerns

OpenAI is making real-time changes to the rollout of its latest artificial intelligence model, GPT-5, after initial user feedback pointed to performance inconsistencies in ChatGPT. The company confirmed that adjustments are being implemented to improve accuracy, usability, and personalisation features.

Rollout changes and technical fixes

According to OpenAI, the GPT-5 launch introduced an automated model-routing system designed to dynamically select the most appropriate model for a query. However, users reported that the feature occasionally returned responses of lower quality compared to previous models. Chief executive officer Sam Altman acknowledged the issue, stating that the “auto-switcher broke,” which affected perceived performance.

In response, OpenAI has temporarily restored access to the earlier GPT-4o model for ChatGPT Plus subscribers while refining the routing system. The company said it aims to ensure GPT-5 consistently delivers the intended improvements in reasoning, creativity, and contextual understanding.

Expanded usage limits for ChatGPT Plus users

To address user concerns about productivity constraints, OpenAI has raised the usage limits for GPT-5’s advanced “Thinking” mode. ChatGPT Plus subscribers can now send up to 3,000 reasoning-intensive messages per week, significantly higher than during the initial rollout phase. The company indicated that further increases are planned across all subscription tiers, potentially surpassing pre-GPT-5 message allowances.

New personalisation features in ChatGPT

OpenAI is also introducing expanded personalisation controls in ChatGPT. Users can now adjust tone, warmth, and the use of emojis in responses, allowing for a more tailored conversational experience. These settings are intended to replicate some of the stylistic flexibility of GPT-4o, which many users had preferred for its conversational tone.

The update follows the addition of new “personality modes” such as Cynic, Robot, Listener, and Nerd, alongside improvements in voice interaction and upcoming integrations with third-party applications including Google Calendar and Gmail.

Mixed user reception and ongoing updates

While GPT-5 has demonstrated enhanced performance on coding and reasoning benchmarks, feedback from some early adopters has been critical. Reports on online forums and social media platforms highlighted occasional inaccuracies and a perceived drop in creative output.

OpenAI has emphasised that the rollout process is iterative, with engineering teams deploying frequent updates based on real-world user data. The company continues to make GPT-5 available to free-tier ChatGPT users in a limited capacity, while Plus and Pro subscribers receive broader access, additional features, and higher usage thresholds.

For developers, GPT-5 is now available via API in multiple configurations—standard, mini, and nano—allowing integration into applications with varying performance and cost requirements. Advanced parameters such as “reasoning_effort” and “verbosity” provide greater control over output.

OpenAI has not announced a fixed timeline for completing the GPT-5 rollout but stated that enhancements will continue “in the coming weeks” to address technical and user experience concerns.