# GPT 5.4 leaks

Source: https://www.youtube.com/watch?v=jFIBotuTZT4
Recap page: https://rapidrecap.app/video/jFIBotuTZT4
Generated: 2026-03-05T07:33:32.297+00:00

---
## Quick Overview

The speaker confirms widespread rumors that OpenAI's GPT-5.4 model is coming soon with a significantly larger context window (2 million tokens) and improved reasoning abilities, while noting that Anthropic's Claude is currently leading in mobile app downloads due to user backlash against OpenAI's Department of War contract.

**Key Points:**
- Rumors confirmed that OpenAI's GPT-5.4 model is launching soon with a 2 million token context window, a major increase from the previous 1 million tokens.
- The new GPT-5.4 model shows significantly better reasoning and is designed to handle long-running tasks without forgetting context.
- Anthropic's Claude mobile app has surpassed ChatGPT in daily US downloads, claiming the #1 spot on the App Store.
- Claude's surge is attributed to user backlash following OpenAI's controversial Department of War contract and subsequent employee departures/deletions of internal messages.
- The speaker references prior leaks suggesting GPT-5.4 would have better reasoning and faster performance than competitors like Google's Gemini models.
- Anthropic CEO's candid comments, including an internal memo attacking OpenAI's 'mendacious' Pentagon announcement, further fueled the move toward Claude.
- The speaker notes that while image processing is supposedly better in the new GPT model, the core focus is on longer context and reasoning improvements.

![Screenshot at 00:20: A screenshot displaying a tweet from Wes Roth detailing OpenAI's controversial Department of War contract, the resulting 770% spike in 1-star reviews for the ChatGPT mobile app, and Claude surpassing ChatGPT in daily US downloads.](https://ss.rapidrecap.app/screens/jFIBotuTZT4/00-00-20.jpg)

**Context:** The video discusses recent developments and rumors surrounding major large language models (LLMs), specifically focusing on the anticipated release of OpenAI's GPT-5.4 and the market reaction following OpenAI's controversy involving a Department of War contract. This controversy has seen a significant surge in popularity for competitor Anthropic and its model, Claude, especially in mobile app adoption.

## Detailed Analysis

The speaker confirms several circulating rumors regarding OpenAI's upcoming GPT-5.4 model, stating it will feature a 2 million token context window, a substantial increase from the 1 million token context window seen in earlier models like GPT-4 and Gemini 3. This larger context window, coupled with confirmed improvements in reasoning (specifically citing a 27% reduction in hallucinations), suggests GPT-5.4 will be much better at handling long-running, complex tasks without forgetting details or context. Simultaneously, the speaker highlights a major shift in user preference, evidenced by Anthropic's Claude mobile app achieving the #1 spot on the App Store, surpassing ChatGPT in daily downloads. This surge is directly linked to the public backlash against OpenAI for securing a controversial Department of War contract, which led to employee resignations and the deletion of internal messages related to safety concerns. The speaker also mentions that Anthropic’s CEO reportedly told an employee not to delete messages regarding safety concerns, and that Anthropic's own models (like Claude 2.1) already outperform some of OpenAI's models in certain benchmarks, suggesting Anthropic’s focus on safety and ethical concerns is resonating with users.

### GPT-5.4 Rumors & Features

- 2 million token context window confirmed
- 27% fewer hallucinations
- better reasoning for long tasks
- 1.5x faster than current models
- similar to Gemini 3 benchmarks but with better safety focus

### Market Reaction & Anthropic's Rise

- Claude mobile app surpassed ChatGPT in US downloads
- driven by backlash over OpenAI's DoD contract
- Anthropic CEO's candid comments and internal memos fueled user migration

### OpenAI's Response & Internal Issues

- OpenAI leadership confirmed to be prioritizing avoiding 'hype' and 'letdown' cycles
- employee screenshots show internal conflict over safety/alignment vs. rapid releases

![Screenshot at 00:20: A screenshot displaying a tweet from Wes Roth detailing OpenAI's controversial Department of War contract, the resulting 770% spike in 1-star reviews for the ChatGPT mobile app, and Claude surpassing ChatGPT in daily US downloads.](https://ss.rapidrecap.app/screens/jFIBotuTZT4/00-00-20.jpg)
![Screenshot at 00:51: A news clipping showing the headline: 'Anthropic CEO's Candid Comments Imperil Chances of a Compromise' alongside another headline referencing the CEO's memo attacking OpenAI's 'Mendacious' Pentagon Announcement.](https://ss.rapidrecap.app/screens/jFIBotuTZT4/00-00-51.jpg)
![Screenshot at 01:05: A news article snippet referencing Anthropic CEO's memo attacking OpenAI's 'Mendacious' Pentagon Announcement, with the speaker expressing strong opinions about the situation.](https://ss.rapidrecap.app/screens/jFIBotuTZT4/00-01-05.jpg)
![Screenshot at 04:06: The speaker discusses the focus of GPT-5.4 being on reasoning improvements rather than just image generation capabilities.](https://ss.rapidrecap.app/screens/jFIBotuTZT4/00-04-06.jpg)
![Screenshot at 08:00: The speaker gestures while discussing the confirmed 2 million token context window for the upcoming GPT-5.4 model.](https://ss.rapidrecap.app/screens/jFIBotuTZT4/00-08-00.jpg)