OpenAI had a very busy week.
First, CEO Sam Altman had a wee freakout over ChatGPTβs loss of market share to Googleβs Gemini 3 and Anthropicβs Claude Opus 4.5, both of which were released in November. Altman issued an internal βcode red.β A few days later, OpenAI moved up its launch of GPT-5.2, which has three modes:
- Instant: I need that answer right now.
- Thinking: I need a plan or a code, but Iβll brew some coffee while I wait for a better answer.
- Pro: I have a really difficult question, so let me take a nap while you do all the legwork for me.
GPT-5.2 now wins most of OpenAIβs own benchmarks, but R&D World ran a side-by-side comparison of six of the most common benchmarks. Here were the winners:
- SWE-bench Verified (coding): Claude Opus 4.5
- GPQA Diamond (science): Gemini 3 Deep Think
- AIME 2025 (math, no tools): GPT-5.2 Thinking/Pro
- ARC-AGI-2 (abstract reasoning): GPT-5.2 Pro
- Humanityβs Last Exam (academic-level reasoning): Gemini 3 Deep Think
- FrontierMath Tier 1-3: GPT-5.2 Thinking
At the moment, Anthropicβs Claude remains the gold standard for coding, which is in line with Product Hunt usersβ opinions. But OpenAI has upped its game on math and reasoning.
OpenAI didnβt just make moves on the LLM front. It also moved forward on AI-generated video. Hereβs what happened:
The same day OpenAI released 5.2, it announced a deal with Disney. The House of Mouse is taking a $1B stake in the company and will βbecome a major customer of OpenAI.β Itβs also licensing over 200 characters from its extensive IP library. That means that Sora users will soon be able to dream up a video of Buzz Lightyear and Woody sumo wrestling in a tub of mayonnaise without running into any legal issues (as long as Hellmanβs doesnβt get involved). Disney+ will even feature some of the videos.
The move obviously injects some life into OpenAI, but it also turns one of the biggest entertainment companies in the world into a surrogate. To wit, the day before the announcement, Disney issued Google a cease-and-desist over copyright infringement on its Gemini 3 model.
Sure, maybe Google will end up signing a similar licensing deal with Disney. But for now, hereβs where things stand on AI video generation:
Sora does text-to-video and image-to-video for up to one minute. It is now included in ChatGPT Plus and Pro subscriptions. Googleβs Veo goes longer than a minute and comes bundled with Gemini subscriptions. But only one lets Disney adults geek out with Goofy.