The closing weeks of 2025 have turned into a direct contest between OpenAI and Google. OpenAI released GPT-5.2 on December 11, only weeks after GPT-5.1, while Google’s Gemini 3, launched in November, continues to draw praise from users for speed and for the image generation of its Nano Banana Pro model. Early reactions suggest the two releases have landed very differently.
OpenAI describes GPT-5.2 as its most advanced model for everyday professional work, citing gains in reasoning, long-context understanding and coding. Many developers and reviewers who have compared it with GPT-5.1 in day-to-day use report improvements that are real but modest, while Gemini 3 has been widely perceived as a more noticeable step forward for ordinary users.
The code red memo
The competitive pressure was visible inside OpenAI before GPT-5.2 shipped. On December 1, chief executive Sam Altman sent an internal memo declaring a “code red,” according to reporting by Fortune, Forbes and other outlets. The memo, as described in those reports, told staff to redirect resources toward concrete, user-visible improvements to ChatGPT and to deprioritize some longer-term projects, after Gemini 3 outperformed GPT-5.1 on several public benchmarks and began winning over users.
GPT-5.2 arrived ten days later. The timing has led many observers to read the release as a direct response to Google, although OpenAI has not characterized it that way.
Price and performance
One clear difference between the two models is cost. OpenAI set GPT-5.2’s API price at $1.75 per million input tokens, up from $1.25 for GPT-5.1, an increase of about 40 percent. Google has positioned Gemini 3 as the lower-cost option, and users comparing the two have generally found Gemini 3 faster and more concise in its answers.
In creative tasks, Gemini 3’s image generation has drawn particular attention. Nano Banana Pro, Google’s image model, produces results for thumbnails, sketches and complex visual concepts that users can judge immediately, without needing benchmark scores to interpret. For reasoning-heavy prompts, reviewers have described Gemini 3’s responses as coherent and quick, while GPT-5.2’s advantages tend to show up in longer, more structured professional work.
What the contest shows
The episode highlights a shift in how AI models are judged. Benchmark gains and version numbers matter less to the mass market than whether a release feels faster, cheaper or more capable in routine use. By that measure, Gemini 3 has been received as a strong release and GPT-5.2 as a cautious one, even though OpenAI’s own figures show gains across its evaluation suite.
Both companies now face pressure to deliver improvements that users can perceive directly, not just better scores. OpenAI’s code red memo suggests its leadership has reached the same conclusion.
