OpenAI's Internal 'Code Red' and Response (Altman Memo / GPT-5 vs Gemini)

OpenAI has issued a 'Code Red' memo, prioritizing the enhancement of ChatGPT's capabilities and stability. The memo warns that OpenAI is facing a survival crisis. Altman outlines five key priorities: improving image generation, enhancing model capabilities and stability, winning on leaderboards, increasing speed and stability, and reducing unnecessary refusals. The memo indicates that a new inference model will be released next week, claiming it will surpass Google Gemini 3. Several commercialization projects have been postponed, including advertising, AI Agents, and Pulse personalized reports. Reports indicate that GPT-5's pre-training has encountered bottlenecks, and the shift to RL scaling has led to a decline in user experience. While GPT-5.1 has fixed some bugs, multiple users and comparative examples point out that its translation and conversation quality are inferior to Gemini 3. The situation is compared to Google's 'Code Red' response to ChatGPT three years ago, suggesting a role reversal.
Key insights
Altman Issues 'Code Red', Outlines Five Priorities and Announces New Model Next Week
The internal memo (Code Red) warns that OpenAI is facing a survival crisis and clearly outlines five priorities: improving image generation, enhancing model strength and stability, winning on various leaderboards (such as LM Arena), increasing ChatGPT's speed and stability, and reducing unwarranted refusals. The memo states that a new inference model will be released next week, with the direct goal of surpassing Google's Gemini 3. This information comes from the disclosure of the internal memo and related reports, listing specific action directions and timelines.
OpenAI Suspends Multiple Commercial Projects to Focus on Development
The memo explicitly states that several commercialization projects have been put on hold, affecting initiatives such as advertising, AI Agents (automated agents for shopping, healthcare, etc.), and Pulse (automated personalized reports). These projects have been postponed to concentrate resources on enhancing ChatGPT's capabilities, speed, and stability, as well as launching the new inference model.
GPT-5 Pre-training Encounters Bottlenecks, User Experience Declines
Multiple posts indicate that GPT-5 has hit a 'wall' during its pre-training phase, with the team shifting to reinforcement learning scaling (RL scaling), resulting in a decline in conversation and daily usage experience for ordinary users. Reports mention that GPT-5 has encountered understanding or expression errors in basic tasks like translation. Although GPT-5.1 has addressed some issues, it is still rated lower than the newly released Gemini by users.
Gemini 3 Achieves Breakthrough in Pre-training, Translation and Conversation Experience Superior to GPT-5
Comparative examples and user tests show that Gemini 3.0 Pro has made improvements in pre-training, with accurate translation results and clear expressions. Under the same tasks, GPT-5 (chat/thinking mode) has been pointed out to have understanding or expression flaws. Multiple posts use translation examples as evidence, emphasizing Gemini's advantages in daily conversation and quality.
Altman Emphasizes Winning on Leaderboards and Speed Stability
The memo lists 'winning on various leaderboards (such as LM Arena)' as one of the priorities, while also emphasizing the need to improve ChatGPT's speed and stability, and to reduce the model's frequent refusals to answer 'harmless questions' to enhance user experience and strive for leading evaluation metrics.
This 'Code Red' is Compared to Google's Response to ChatGPT Three Years Ago
Social media posts compare this internal 'Code Red' at OpenAI to Google's 'Code Red' response to ChatGPT three years ago, suggesting a role reversal: Google was impacted back then, while now OpenAI faces a counterattack from Google's Gemini.