OpenAI Retires GPT-4o, Launches 'Spark' Coder
OpenAI has launched "Spark," a lighter and faster coding assistant aimed at increasing developer efficiency. The company also announced the retirement of its GPT-4o model, which prompted backlash from developers who had built products on the platform. In response to the disruption, some users are calling for a "classic mode" to maintain backward compatibility for deprecated models.
- "Spark," officially named GPT-5.3-Codex-Spark, is designed for real-time, interactive coding and is reportedly up to 15 times faster than the flagship GPT-5.3-Codex. It achieves this speed through technical optimizations, including an 80% reduction in client-server round-trip overhead and a persistent WebSocket connection to maintain an active session. - The new coding assistant runs on specialized hardware from AI chipmaker Cerebras, specifically the Wafer Scale Engine 3, as part of a multi-year partnership with OpenAI to reduce latency. This allows Spark to be used for more conversational "pair programming" tasks, like quick refactors and bug fixes, rather than large, autonomous projects. - The retirement of GPT-4o is scheduled for February 13, 2026, a date some users have criticized as being cruelly close to Valentine's Day, given the emotional attachments many had formed with the model. This is not the first time OpenAI has attempted to retire the model; a similar move in August 2025 was reversed after significant user backlash. - While OpenAI has stated that only 0.1% of daily users were still actively using GPT-4o, this could still represent hundreds of thousands of people. In response to the shutdown, a group called the #Keep4o Movement has organized protests and accused the company of a lack of care. - While the official reason for the retirement is to focus resources on newer models like GPT-5.2, reports suggest that concerns over harmful outcomes also played a role. There are 13 lawsuits that have been consolidated, alleging links between GPT-4o's human-like bonding capabilities and instances of psychotic delusions and suicide attempts. - In response to the shutdown, some users have proposed a "classic mode" that would allow continued access to deprecated models. This would address concerns from developers and users who have built workflows around specific model behaviors and view the removal of a tool they depend on as a breach of trust. - Despite the retirement of GPT-4o from the main ChatGPT interface, it will remain accessible for a longer period for business, enterprise, and education customers, and will continue to be available through the API, with OpenAI promising advance notice before any API retirements. The voice and image features that use a similar base model will not be affected by this change.
