OpenAI Launches GPT-5.4 Days Just After Last Version as 'QuitGPT' Exodus Gains Steam

OpenAI has released its latest model, GPT-5.4, on Thursday, coinciding with a significant user backlash known as the QuitGPT movement. This movement, which has seen an estimated 2.5 million users take action against the company, was triggered by OpenAI's acceptance of a Pentagon AI contract that Anthropic had refused due to concerns over autonomous weapons and mass surveillance. CEO Sam Altman is reportedly seeking to regain user trust.

GPT-5.4 Features and Performance

GPT-5.4 introduces enhanced reasoning, coding, and agentic capabilities, along with a substantial 1-million-token context window. This allows users to handle larger amounts of information in a single session. Benchmarks indicate significant improvements over previous versions:

• •
  On the GDPval benchmark, GPT-5.4 matches or surpasses industry professionals in 83.0% of comparisons, an increase from GPT-5.2's 70.9%.
• •
  In computer use tests (OSWorld-Verified), GPT-5.4 achieved a 75.0% success rate, surpassing the human baseline of 72.4% and significantly outperforming GPT-5.2's 47.3%.
• •
  Performance on deep web research (BrowseComp) also saw a 17 percentage point jump over GPT-5.2.

A notable feature is the mid-response steering, which allows users to redirect the model during its processing, saving time and computation by avoiding the need to regenerate tokens after an error.

Impact on Different User Groups

While benchmarks show broad improvements, the actual gains for users of GPT-5.3 may be more incremental. The model is also noted for being more token-efficient than GPT-5.2, which is particularly beneficial for enterprise users and developers utilizing the API, as it can lead to reduced costs.

However, the full capabilities of GPT-5.4, especially those requiring extensive reasoning, are not yet available to all users. Some advanced features were tested against GPT-5.2, skipping GPT-5.3, and the model's rollout is gradual, with advanced reasoning features still on GPT-5.2 for some users.

Enterprise users are expected to benefit the most. Internal benchmarks show GPT-5.4 scoring 87.3% on a spreadsheet modeling task compared to GPT-5.2's 68.4%. Legal research firm Harvey reported a 91% score on its BigLaw Bench eval. Companies like Mainstay have reported significantly faster sessions and reduced token usage when running agents across property tax portals.