The AI wars are in full swing lately. First, Google dropped Gemini 3 Pro, followed closely by the release of Claude Opus 4.5. Both models took the lead over the previously released GPT-5.1 in both math and science benchmarks. In response, OpenAI brought their new model, GPT-5.2, to the table on December 11, 2025 (New York Time). Its performance in various comprehensive tests is significantly superior to its two predecessors.
Plus and Pro users get immediate access, while Free and "Go" subscription users will gain limited access the following day.
As you can see, GPT-5.2 Thinking outperforms the competing Opus 4.5 and Gemini 3 Pro in math and science tests. Furthermore, it has received enhancements in emotional intelligence, reasoning capabilities, and context handling. I specifically tested its coding abilities; when paired with CodeX connected to a repository, I found its code to be much more concise than Opus 4.5's while achieving the exact same functionality. This proves that its coding prowess has surpassed Opus 4.5, which was previously known for its ultra-strong coding capabilities.
Additionally, starting with GPT-5.1, ChatGPT has allowed users to manually toggle "Thinking Mode" and request extra thinking time (as shown above). This can provide deeper and more accurate answers. However, the downside is obvious: although GPT-5.2 with extra thinking time beats Gemini 3 and Opus 4.5 and is clearly better, the response time has become incredibly long—many questions take 3 to 5 minutes to generate an answer.
I have to say, this year has been the fastest-growing year for the AI industry. Domestic cloud computing vendors in China are fighting a fierce battle, even launching "AI Phones." Internationally, major tech giants are also entering the arena, driving rapid model iteration and growth through constant competition. This "cash-burning" war allows us to use various AIs that are evolving daily with different specialties. In this new era, I hope we can all seize the opportunities and win the future together.
Comments NOTHING