In order to correspond Google launched Gemini 3 Pro, previously the market had already heard OpenAI will release a more powerful new model, and just as the rumors suggested, OpenAI has now officially unveiled the new GPT-5.2 Not surprisingly, the model beat Gemini 3 Pro on every benchmark. The only downside is that it didn’t get a simultaneous update.ImageGeneration capability. In addition, according to foreign media reports, OpenAI has also announced the timing for the adult mode to debut. Here’s a summary for everyone.

OpenAI’s all-new GPT-5.2 model debuts, rolling out to paid users starting now.
OpenAI’s latest GPT-5.2 model shows noticeable improvements in all aspects, not only faster, but also better in data search, presentation creation, writing, and translation. Just like the previous generation, GPT-5.2 in ChatGPT also has three modes, namely
- GPT‑5.2 Instant: Fastest speed, with significant improvements in information lookup, how-to guides, step-by-step instructions, technical writing, and translation, while continuing the warm conversational tone introduced in GPT‑5.1 Instant.
- GPT‑5.2 Thinking: Designed for deeper work, with clearer response structure and more practical details.
- GPT‑5.2 Pro: Currently the smartest, most reliable model, suited for difficult questions that require high-quality answers and can tolerate longer response times.
The figure below compares GPT-5.2 Thinking with the previous generation GPT-5.1 Thinking. As shown, in the most commonly used knowledge work tasks, the improvement exceeds 80%.
| Benchmark test | Field | GPT-5.2 Thinking | GPT-5.1 Thinking | degree of improvement |
|---|---|---|---|---|
| GDPval (win or tie) | Knowledge work tasks | 70.9% | 38.8% | +82.7% |
| SWE-Bench Pro (Public) | Software engineering | 55.6% | 50.8% | +9.4% |
| SWE-bench Verified | Software engineering | 80.0% | 76.3% | +4.8% |
| GPQA Diamond (no tools) | Scientific question | 92.4% | 88.1% | +4.9% |
| CharXiv Reasoning (Python Assisted) | Scientific Chart Problem | 88.7% | 80.3% | +10.5% |
| HMMT (February 2025) | Math competition | 99.4% | 96.3% | +3.2% |
| FrontierMath (Levels 1 to 3) | Advanced Mathematics | 40.3% | 31.0% | +30.0% |
| ARC-AGI-1 (Verified) | Abstract reasoning | 86.2% | 72.8% | +18.4% |
| ARC-AGI-2 (Verified) | Abstract reasoning | 52.9% | 17.6% | +200.6% |
As for the hallucination problem that everyone cares about, GPT-5.2 Thinking also improved, with an error rate of only 6.2%, down 38% from GPT-5.1 Thinking’s 8.8%:

GPT-5.2 Thinking is also OpenAI’s most powerful vision model to date, with error rates in chart reasoning and software interface understanding cut by about half. This means you can worry less about AI making mistakes in recognition and judgment.
Science chart questions improved by 8.4%, accuracy reaching 88.7%:

The improvement in graphical user interface screenshot understanding is even greater, at 22.1%, and GPT-5.2 Thinking’s accuracy now exceeds 86%.

And compared to Google’s most powerful Gemini 3 Pro? OpenAI also provided comparison test data in its official X community, with GPT-5.2 Thinking taking the lead across the board:

OpenAI says GPT‑5.2 Instant, Thinking, and Pro will begin rolling out in ChatGPT starting today, available first to paid plan users (Plus, Pro, Go, Business, Enterprise). If you don’t see it yet, try again later. The previous generation, GPT‑5.1, will also remain available to paid users in its current form and will be officially discontinued after three months.
ChatGPT adult mode launch timeline plan
Besides unveiling GPT-5.2, according to The Verge, during an earlier briefing OpenAI’s head of applications, Fidji Simo, also told reporters that ChatGPT’s “adult mode” is expected to launch in the first quarter of 2026. She added that before the official release, OpenAI hopes to make “age prediction” more accurate.
Additionally, Sam Altman also teased in the community that there will be a small Christmas gift next week, without revealing details. However, a few days ago, the market saw reports that ChatGPT is testing a “holiday offer” that gives away 3 months of ChatGPT Plus membership for free—not sure if that’s the one:
Source: KOCPC Chinese