OpenAI Declares 'Code Red,' Prioritizes ChatGPT Improvements Amid Competition
Serge Bulaev
OpenAI has called a 'Code Red' to urgently improve ChatGPT after new competition from Google's Gemini 3. The company paused smaller projects and is now working fast to make ChatGPT quicker, more reliable, and better at reasoning. In just twelve days, OpenAI released GPT-5.2 with smarter answers and
OpenAI has initiated a 'Code Red' to accelerate ChatGPT improvements, a direct response to intensifying competition from Google's new Gemini 3 model. According to an internal memo cited by Axios, CEO Sam Altman has pushed the company into a high-stakes development sprint, pausing other projects to enhance ChatGPT's speed, reliability, and reasoning capabilities.
A Six-Week Sprint to Refine ChatGPT
The 'Code Red' is an emergency protocol reallocating company resources into a focused, short-term push to upgrade a core product. For ChatGPT, this means an intense focus on six key deliverables designed to counter competitive threats, including performance boosts, better personalization, and a new reasoning-centric model variant.
The directive, detailed in a memo reviewed by Gary Marcus, outlines six core deliverables for the sprint:
- Personalization features that remember user preferences
- Improved benchmark scores on LM Arena and SWE-bench
- Reduced latency for sub-second response times
- Fewer over-refusals on sensitive queries
- A reasoning-focused model nicknamed "ChatGPT Garlic"
- Upgraded image generation via DALL-E
This aggressive timeline has already yielded results, with OpenAI shipping GPT-5.2 just twelve days into the sprint. Reports suggest these sprints, which have occurred "once or twice a year," typically last six to eight weeks, placing the end of the current push in early February 2026.
How ChatGPT Stacks Up Against Gemini 3
The urgency of the 'Code Red' is underscored by recent benchmarks where Google's new model shows an advantage in key areas, particularly multimodal understanding and factual accuracy. However, ChatGPT maintains a critical lead in coding-related tasks.
| Metric | Gemini 3 Flash | GPT-5.2 Extra High |
|---|---|---|
| MMMU-Pro (multimodal) | 81.2 percent | 79.5 percent |
| SimpleQA Verified | 68.7 percent | 38.0 percent |
| Video-MMMU | 86.9 percent | 85.9 percent |
| SWE-bench (coding) | 78.0 percent | 80.0 percent |
While Gemini 3's native multimodality and lower token costs threaten to lure developers, ChatGPT's continued dominance in coding tasks is vital for enterprise adoption. This aligns with what analysts call OpenAI's strategy to become an "operating system for the AI age" by focusing on production-ready infrastructure over pure research.
Although the current crunch has delayed some features like an "adult mode" to Q1 2026, the success of this sprint will show whether such rapid release cadences become the new industry standard.
What triggered OpenAI's "Code Red" declaration in December 2025?
CEO Sam Altman issued the alert after Google released Gemini 3, a model that beats ChatGPT 5.2 on four headline benchmarks: MMMU-Pro multimodal reasoning (81.2 % vs 79.5 %), multilingual MMMLU (91.8 % vs 89.6 %), SimpleQA factual accuracy (68.7 % vs 38.0 %) and video understanding (86.9 % vs 85.9 %). The speed of the launch - combined with Gemini 3 Flash pricing as low as $0.50 per million tokens - convinced OpenAI leadership that rapid counter-measures were required to protect user and developer mind-share.
How is the company reorganising resources under the Code Red?
Engineering and product teams have been consolidated into a single ChatGPT strike-force. Six priority tracks were leaked in an internal memo:
1. Personalisation & custom instructions
2. LM Arena score gains
3. Latency & uptime
4. Fewer over-refusals
5. A new reasoning model dubbed "ChatGPT Garlic"
6. Image-generation polish
Longer-term research strands that lack a clear 90-day path to the consumer product have been paused or deprioritised, signalling a temporary swing from exploration to pure execution.
Will release cadence really speed up and, if so, for how long?
Yes - the first proof came on 12 December 2025, when GPT-5.2 shipped only three weeks after GPT-5.1, breaking OpenAI's normal monthly rhythm. Altman told staff that Code Red phases historically last "six or eight weeks", implying the current sprint should wind down near the end of January 2026. He also warned investors to expect one or two such emergencies every year as competition heats up, so shorter cycles may become routine rather than exceptional.
Where does ChatGPT still outperform Gemini 3?
Independent late-2025 tests give ChatGPT the edge in:
- Coding accuracy - 80.0 % on SWE-bench versus Gemini 3 Flash's 78.0 %
- Developer ecosystem - deeper integration with Microsoft IDEs, GitHub Copilot and enterprise Azure endpoints
- Token efficiency for repeat queries - 90 % input-cache discounts make high-volume API usage cheaper for engineering teams
These strengths explain why OpenAI is doubling down on "everyday professional applications" instead of trying to match Google on multimodal flair alone.
What could this strategic pivot mean for users and the wider AI field?
Short term, customers should see:
- Noticeably faster response times and fewer "I can't answer that" refusals
- Tighter integration with spreadsheets, slide decks and long documents
- A forthcoming "adult mode" toggle, now slated for Q1 2026
Analysts at eMarketer frame the shift as OpenAI's move "from model maker to operating infrastructure", betting that developer loyalty and enterprise reliability will matter more than headline benchmark crowns. If the gamble works, competitors may be forced to follow suit, accelerating the entire industry's transition from research demos to production-ready utility.