# GPT-5.2 is Here

Source: https://www.youtube.com/watch?v=_6WTIDY7aYs
Recap page: https://rapidrecap.app/video/_6WTIDY7aYs
Generated: 2025-12-12T02:02:21.224+00:00

---
## Quick Overview

GPT-5.2 is a significant step forward, especially in instruction-following and code generation, but it is notably slower than GPT-5.1, with its Pro version being too slow for daily use and its Thinking mode being slow for most questions, although it excels in deep reasoning tasks and shows major improvements in visual and long-context understanding.

**Key Points:**
- GPT-5.2 Thinking is a meaningful step forward in instruction-following and complex task execution, particularly code generation, which is noted as significantly better than GPT-5.1.
- The Pro version of GPT-5.2 is considered incredibly impressive for deep reasoning but is too slow for daily use, often resulting in long thinking times that waste time.
- Performance benchmarks show GPT-5.2 Thinking beating or tying industry professionals on 70.9% of GDPVal knowledge work tasks, outperforming GPT-5.1 significantly in areas like ARC-AGI-2 (52.9% vs 17.8%).
- The model demonstrated a 390X efficiency improvement in one year on the ARC-AGI-1 task when comparing GPT-5.2 Pro SOTA score (90.5% at $11.64/task) to a previous version (88% at $4.5/task).
- Early feedback suggests GPT-5.2 excels in professional deliverables like spreadsheets and presentations, with the Pro version being more polished and deliberate than the chaotic freelancing style of GPT-5.1.
- The Disney/OpenAI partnership was announced, granting Sora access to Disney characters for user-prompted video generation, with Disney also investing $1 billion in OpenAI.
- A key downside noted is the speed; GPT-5.2 Thinking is slow for most questions, and Pro is often too slow for real-time use, suggesting it's best reserved for complex, deep reasoning tasks.

![Screenshot at 00:07: The OpenAI announcement graphic highlights GPT-5.2 as the most advanced frontier model for professional work and long-running agents, setting the stage for the discussion on its capabilities versus GPT-5.1.](https://ss.rapidrecap.app/screens/_6WTIDY7aYs/00-00-07.png)

**Context:** The video aggregates reactions and initial reviews from various AI researchers and industry figures regarding the release of OpenAI's new model, GPT-5.2, which was announced alongside a major partnership with The Walt Disney Company for their Sora video generation model. The discussion centers on the improvements GPT-5.2 offers over its predecessor, GPT-5.1, particularly in professional tasks, reasoning, and coding, while also highlighting significant trade-offs, primarily in speed, especially with the more capable Pro version.

## Detailed Analysis

The discussion around GPT-5.2 reveals a consensus that it represents a meaningful, but incremental, upgrade over GPT-5.1, heavily favoring professional use cases. The Thinking mode shows significant improvements in instruction-following, code generation (notably better than 5.1), and vision/long-context understanding, as evidenced by the ARC-AGI benchmark where 5.2 scores 88% compared to 5.1's 72% on one metric, and a 390X efficiency improvement was noted on the ARC-AGI-1 task. However, speed is a major drawback; the Thinking mode is described as very slow for most questions, leading users to skip it for instant results. The Pro version is praised for deep reasoning and a more polished, professional writing style, outperforming 5.1 in client-facing scenarios, but it is too slow for daily tasks, sometimes taking an absurdly long time on research tasks. The release was accompanied by a major partnership with Disney to bring characters into Sora, an event that seemed to overshadow the technical release for some. Overall, GPT-5.2 is seen as a worthwhile upgrade for complex, professional work, but users must manage expectations regarding speed.

### Performance Benchmarks

- GPT-5.2 Thinking achieves 70.9% on GDPVal vs 38.8% for GPT-5.1
- GPT-5.2 scores 55.6% on SWE-Bench Pro vs 50.8% for GPT-5.1
- GPT-5.2 Thinking scores 92.4% on GPQA Diamond vs 88.1% for GPT-5.1

### Hallucination & Reasoning

- GPT-5.2 Thinking shows a reduction in hallucination rates compared to 5.1, with one user noting 30-40% less hallucination overall
- GPT-5.2 Pro shows superior deep reasoning capabilities compared to GPT-5.1

### Speed & Usability

- GPT-5.2 Thinking is described as slow for most questions, leading some to avoid it for instant answers
- GPT-5.2 Pro is described as incredibly smart but too slow for daily use, often taking an absurdly long time on research tasks

### Professional Deliverables

- GPT-5.2 excels at creating spreadsheets, presentations, and coding, producing cleaner outputs than 5.1
- GPT-5.2 is better at following instructions and less prone to hallucinating UI elements compared to 5.1

### Real-World Feedback

- Early testers noted deeper explanations and richer idea exploration in 5.2 compared to 4.5
- One user noted 5.2's writing style is more polished and professional than 5.1's 'chaotic freelancer' style

### Disney Partnership

- OpenAI and Disney reached a 3-year licensing agreement for Sora to generate videos using Disney characters, with Disney also investing $1 billion in OpenAI

### Overall Takeaway

- GPT-5.2 is a meaningful and noticeable improvement for enterprise use cases, but the speed trade-off is significant, making Pro slow for casual use.

![Screenshot at 00:00: Presenter discussing the launch of GPT-5.2.](https://ss.rapidrecap.app/screens/_6WTIDY7aYs/00-00-00.png)
![Screenshot at 01:00: A comparison table showing GPT-5.2 Thinking outperforming GPT-5.1 Thinking across multiple benchmarks like GDPVal and SWE Bench.](https://ss.rapidrecap.app/screens/_6WTIDY7aYs/00-01-00.png)
![Screenshot at 04:47: A section comparing GPT-5.2 Thinking and GPT-5.1 Thinking on SWE-Bench Pro, showing 5.2 achieving a state-of-the-art score of 55.6%.](https://ss.rapidrecap.app/screens/_6WTIDY7aYs/00-04-47.png)
![Screenshot at 07:53: A review article titled 'GPT-5.2 vs. GPT-5.1: What Actually Changed?' highlighting GPT-5.2 as the flagship model.](https://ss.rapidrecap.app/screens/_6WTIDY7aYs/00-07-53.png)
![Screenshot at 15:50: A Twitter post from ARC Prize showing a graph where GPT-5.2 Pro achieves a 390X efficiency improvement on the ARC-AGI-1 task compared to a previous run.](https://ss.rapidrecap.app/screens/_6WTIDY7aYs/00-15-50.png)