GPT 5.4 Leak: OpenAI Employee's Slip Reveals the Next Model
An OpenAI Codex team member's Twitter slip revealed the existence of 'GPT 5.4.' Additional evidence found on Reddit and ChatGPT's backend makes a release this week highly likely. With DeepSeek V4 also expected the same week, a US-China AI showdown is imminent.
OpenAI's next model, GPT 5.4, was revealed to the world through an employee's mistake. A member of the Codex team posted 'GPT-5.4' on Twitter, then quickly deleted it and replaced it with '5.3 Spark,' but screenshots had already been captured. Subsequent discoveries on Reddit and ChatGPT's backend have made GPT 5.4's existence virtually confirmed.
1. How GPT 5.4 Was Leaked: From a Twitter Slip to Backend Evidence
It started with a simple mistake. An OpenAI Codex team member posted a tweet mentioning 'GPT-5.4,' only to delete it within minutes and replace the reference with '5.3 Spark.' But the internet never forgets. Screenshots spread instantly.
The smoking gun came from the tech community. Reddit users discovered 'alpha-gpt-5.4' in OpenAI's public API /models endpoint. ChatGPT's web backend also revealed routing identifiers including gpt-5-4, gpt-5-4-instant, gpt-5-4-thinking, and gpt-5-4-pro. This wasn't mere rumor; it was a model actually registered in the production system.
2. GPT Numbering History: Sprinting at Two-Month Intervals
OpenAI's GPT 5 series has evolved at a remarkable pace. Starting with GPT 5.0 in August 2025, followed by 5.1 in November, 5.2 in January 2026, and the Codex-exclusive 5.3 on February 5. That's roughly a two-to-three-month iteration cycle.
Notably, 5.3 was released exclusively for Codex, OpenAI's coding agent. General ChatGPT users never experienced 5.3 and will likely go straight to 5.4. The variant identifiers discovered in the backend, such as gpt-5-4-thinking and gpt-5-4-pro, suggest GPT 5.4 will bring a substantial performance leap. The 'thinking' variant points to enhanced reasoning capabilities, while 'pro' hints at a high-performance expert mode. Historical patterns show that once model routing identifiers appear in the backend, official release typically follows within about a week. Some observers predict a launch as early as February 26.
3. DeepSeek V4 Showdown: A US-China AI Head-to-Head
What makes GPT 5.4's timing especially noteworthy is the coincidence. DeepSeek has also announced its next model, V4, for the same week. With projections suggesting DeepSeek V4 could surpass Claude and ChatGPT on coding benchmarks, a head-to-head showdown between the latest US and Chinese AI models is taking shape.
This rapid release cadence isn't just about technological competition. Considering the massive distillation attacks by Chinese AI firms that Anthropic and OpenAI recently exposed, rapid model replacement itself serves as a defense strategy against distillation. When models are quickly superseded, the value of distilled copies of older versions plummets.
Closing: Can GPT 5.4 Rescue an OpenAI Under Siege?
GPT 5.4 is no longer a secret. Triple evidence from a Twitter slip, API endpoints, and backend routing has been secured, and the thinking and pro variants suggest a major performance upgrade is on the horizon.
But for OpenAI, GPT 5.4 carries significance far beyond a routine model release. Besieged by massive distillation attacks from Chinese AI firms, fierce competition from Anthropic and Google, and internal talent departures and organizational turmoil, OpenAI finds itself cornered on multiple fronts. In this context, GPT 5.4 is a critical card to reassert technological supremacy and restore market confidence.
The head-to-head with DeepSeek V4, slated for the same week, will serve as the ultimate proving ground. With DeepSeek V4 projecting it will surpass existing models on coding benchmarks, all eyes are on whether GPT 5.4 can demonstrate a decisive edge.
