OpenAI’s “code red” was a reported internal push to improve ChatGPT—not a public emergency declaration or proof that Google had won the AI race. On December 1, 2025, Sam Altman reportedly told employees to focus on ChatGPT as Google’s Gemini 3 gained attention. OpenAI announced GPT-5.2 ten days later, but there is no public metric showing that the internal effort itself improved market share or user retention. The episode is best understood as a strategic response, not a verdict on which assistant is better today.
What OpenAI’s reported “code red” meant
The Information and The Wall Street Journal reported that Altman issued the directive on December 1, 2025; wider coverage followed on December 2. The account was based on an internal memo, which was not released publicly in full. The Associated Press also summarized the report. OpenAI did not initially publish a full public confirmation of the memo. The Information’s report and the AP account are therefore evidence of what was reported, not a complete public record of the directive.
In this context, “code red” described an internal management priority: concentrate people and resources on ChatGPT’s quality, personalization, speed, and reliability, rather than spread effort across as many initiatives. It was not a regulatory classification, public safety alert, or announcement that OpenAI was about to fail.
Projects reportedly put on hold or deprioritized
Coverage of the memo described advertising-related work, shopping and health agents, the Pulse personalized reporting feature, and other projects as delayed or receiving less attention. Those specifics remain attributed to reporting because the complete memo is not public. The practical implication is that OpenAI was reportedly weighing near-term improvements to its flagship chatbot against other product and monetization bets.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →#1 Best Overall
Why Gemini 3 changed the competitive story
Google’s Gemini 3 arrived in November 2025 and drew attention for strong results on several reported industry benchmarks. Coverage also described renewed interest from consumers and businesses. That helped make Gemini a credible competitive concern, but a high score on selected tests does not establish that a model is better for every task or that users prefer its product. Axios’s account of the pressure on AI rivals and The Atlantic’s analysis describe the broader context.
Google’s advantage is not only a model. Gemini can be connected to products people already use, including Search, Android, Workspace, and Google Cloud. Google also has infrastructure, data, and custom AI chips. That distribution could put a capable assistant in front of users without requiring them to adopt a separate chatbot, and it matters to enterprise buyers considering AI within existing systems. Distribution does not guarantee superior model quality, privacy terms, or user satisfaction, but it can make adoption easier.
OpenAI, meanwhile, had a large ChatGPT user base and a growing range of products, but broad expansion can compete for engineering attention. The challenge was also larger than a two-company contest: contemporary reporting identified Anthropic as a serious rival, especially in enterprise and coding. Operating frontier models at scale and finding sustainable ways to monetize consumer use added pressure. Fortune’s follow-up discussed the wider competitive picture, including enterprise-market estimates that depend on the datasets and definitions used.
Does “Gemini beat ChatGPT” hold up?
Only if the claim names a specific test, product version, and date. “ChatGPT” and “Gemini” are services as well as models: their interfaces, search, integrations, available tools, limits, and plans affect the experience. A benchmark result can help compare a defined capability; it cannot settle which product is best overall.
| Comparison area | What to check |
|---|---|
| Everyday chat and writing | Try representative prompts from your own work. Compare usefulness, clarity, factual errors, and how much editing the answer needs. |
| Search and freshness | Check whether answers use the web, show useful citations, and distinguish sourced facts from inference. Compare response time as well as answer quality. |
| Coding | Separate chatbot assistance from an IDE agent or API model. Test on the same repository, tools, instructions, and task; track successful completion and correction time. |
| Long documents | Check the actual model and product limits, but also test whether it retrieves details accurately from your documents. A large context limit does not guarantee reliable recall. |
| Images, audio, video, and documents | Compare the specific input and output modes available in the plan or product you would use, rather than assuming every model variant supports the same features. |
| Enterprise use | Review administration, integrations, privacy, retention, compliance, and data-location terms alongside capability. |
| Price and limits | Compare equivalent free or paid tiers, usage allowances, included tools, and restrictions. Do not treat a chatbot subscription and API access as the same purchase. |
| Reliability | Repeat tasks that matter to you. A striking demo or isolated failure is not enough to establish consistent performance. |
For benchmark claims, look for the model version, test date, evaluator, prompt method, tool access, and whether the figure is an average, best-of result, or pass rate. Also ask whether the benchmark might be contaminated by training data or vulnerable to overfitting. Without those details, a ranking is a weak basis for a purchase decision.
OpenAI’s response: GPT-5.2
OpenAI announced GPT-5.2 on December 11, 2025, describing Instant, Thinking, and Pro variants. The timing made it widely understood as a response to Gemini 3’s momentum, but timing alone does not prove that the internal directive caused a particular release or technical result. OpenAI said ChatGPT subscription pricing would remain unchanged at launch. OpenAI’s announcement reported a 55.6% SWE-Bench Pro result for GPT-5.2 Thinking, along with results in other areas. These are OpenAI’s reported figures, not independent validation or a universal measure of coding ability.
OpenAI listed these API prices at GPT-5.2’s launch. They are historical launch prices, not a guarantee of current pricing:
| Model at launch | Input per million tokens | Cached input per million tokens | Output per million tokens |
|---|---|---|---|
| GPT-5.2 | $1.75 | $0.175 | $14 |
| GPT-5.2 Pro | $21 | Not listed by OpenAI in the announcement | $168 |
For developers, token rates are only one part of cost. A useful comparison also measures task success, latency, tool calls, retries, and the human time needed to review or repair output. A more expensive model might require fewer retries on a demanding job; a less expensive one may be sufficient for routine extraction or classification.
Timeline: from Gemini 3 to the next OpenAI model
- November 2025: Google released Gemini 3, which drew attention for reported benchmark performance and product momentum.
- December 1, 2025: Altman reportedly directed OpenAI staff to prioritize ChatGPT in an internal “code red.”
- December 2, 2025: News coverage made the reported directive public.
- December 11, 2025: OpenAI announced GPT-5.2.
- December 17, 2025: Fortune reported that Altman expected the effort to last roughly six to eight weeks. This was a reported expectation, not a public performance milestone.
- March 2026: OpenAI announced GPT-5.4, which replaced GPT-5.2 Thinking for some ChatGPT users. That subsequent product change shows the model race continued; it does not establish whether the earlier internal effort succeeded. See OpenAI’s GPT-5.4 announcement.
Did the code red work?
The clearest observable outcome was the GPT-5.2 launch shortly after the directive was reported. The public record summarized here does not provide a causal measure showing that the directive improved ChatGPT’s retention, market share, reliability, or revenue. Nor does a fast product release by itself show that the internal reallocation worked: answering that would require a defined before-and-after metric and a way to separate the effect of the directive from other changes.
The episode was also a reversal in the public narrative. After ChatGPT’s December 2022 launch, Google executives reportedly declared their own “code red” over the perceived threat to Search. In 2025, OpenAI was the company reported to be urgently responding to Google’s progress. That reversal illustrates how quickly perceived leadership can shift; it does not show that one company permanently won. The Guardian’s report covers the 2025 episode and its historical comparison.
What this means for users and buyers
The December 2025 news is not, by itself, a reason to switch assistants or pay for a subscription. Choose by the work you need done and the ecosystem you already use. Try the relevant free tier first when casual use is enough, and check live plan terms before paying: names, prices, included tools, and usage limits can change.
- ChatGPT may suit you if your work depends on OpenAI-specific workflows such as custom GPTs, projects, deep research, Codex, or an established ChatGPT history. Check the current ChatGPT plans and Plus plan details rather than relying on old prices.
- Gemini may suit you if you rely on Gmail, Docs, Drive, Search, Android, or Google’s storage ecosystem. Google’s AI plans and Google One plans describe current bundles; check the live page for the price and limits in your region.
- Claude is worth evaluating if your priorities include writing, long documents, coding, or Claude Code. Compare the relevant product and plan terms on Anthropic’s pricing page and its plan guidance; it is a serious alternative, not a universal winner.
- For coding teams, compare agents separately from chatbots. Test tools such as Codex and Claude Code against your actual repository, permissions, review process, and cost of failed or repeated tasks.
- For businesses, prioritize governance as well as output quality. Review retention, training use, administrative controls, compliance, integrations, and data-location terms before submitting confidential information.
For consequential decisions, cross-check an AI answer against dependable sources or another model; two agreeing answers are not proof of correctness. Avoid uploading sensitive personal or business data until you understand the product’s data terms.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
