GPT-5.2
OpenAI's December 2025 GPT release, offered in three modes plus a coding variant.
GPT-5.2 is a large language model family from OpenAI that arrived on 11 December 2025, taking over from GPT-5.1. It ships as three modes, two of which reason before answering, and it drew attention for knowledge-work tasks and for a reported rush to market after Google's Gemini 3 Pro.
- Released 11 December 2025 as three modes plus a coding variant
- Thinking and Pro reason first; Pro uses more compute per request
- Pro offers standard and extended thinking settings
- Box reported a seven-point gain over GPT-5.1 on knowledge-work tests
- Press linked the timing to an internal "Code Red" after Gemini 3 Pro
What it is
GPT-5.2 is a large language model put out by OpenAI on 11 December 2025. Rather than one single system, it is a set of three: an instant mode built for quick replies, a thinking mode, and a Pro mode. The latter two think before they answer and count as reasoning models, with Pro spending more time and compute on each task. A coding-focused sibling, GPT-5.2-Codex, was released alongside it. The model sits inside the GPT line and follows GPT-5.1. OpenAI describes it as multimodal and as a foundation model. It is offered through ChatGPT and Microsoft Copilot.
How it came about
The launch landed roughly three weeks after Google shipped Gemini 3 Pro, which press coverage at the time treated as the strongest multimodal system available. Reports said an internal memo nicknamed "Code Red", triggered by that position, pushed OpenAI to move the release forward from a late-December plan. Company executives rejected the idea that the launch was purely a reaction. Fidji Simo, OpenAI's CEO of Applications, told Wired that work on the release had been under way for months, while conceding that the memo and extra staff put on ChatGPT helped finish the rollout.
Inside the model
The line is split into three modes with different jobs. Instant is tuned for speed and efficiency. Thinking and Pro both reason first, and Pro draws on a larger share of reasoning time and compute. Pro itself offers two settings, standard and extended, which differ in how much processing power and time each request receives. OpenAI categorises the model as multimodal, a large language model, a generative pre-trained transformer and a foundation model. No parameter count, training corpus, context window or chip details were published with the release.
What it can do
OpenAI highlights gains in building spreadsheets, financial modelling, presentations and projects that need many steps carried out in order. Aaron Levie, who runs Box, said the model came out seven points ahead of GPT-5.1 on his company's tests of real-world knowledge work. VentureBeat observed that early take-up centred on heavy users, developers and enterprise agents rather than everyday conversation. The Guardian found the system citing Grokipedia, an encyclopedia linked to Elon Musk; Nina Jankowicz has criticised that project for weak sourcing.
How to use it
The model is proprietary, so its weights are not published. It reaches people through ChatGPT and Microsoft Copilot, and OpenAI maintains a product page for it. Pro users get a choice between the two thinking settings shown in OpenAI's own screenshots of the interface.
Reception and debate
VentureBeat read early usage as skewed towards power users, developers and enterprise automation instead of casual chat. Box's Aaron Levie measured a seven-point improvement over GPT-5.1 on internal knowledge-work tests. The Guardian reported that the model drew on Grokipedia as a reference, drawing criticism because that encyclopedia has been called poorly sourced by Nina Jankowicz.
Why it matters
The release sits between GPT-5.1 and the models that came next: GPT-5.3-Codex arrived on 5 February 2026, followed by GPT-5.4 on 5 March 2026. Coverage framed it as part of a fast exchange of releases between OpenAI and Google, and as a system aimed at enterprise and developer workloads.
Questions people ask about GPT-5.2
What exactly is GPT-5.2?
It is OpenAI's large language model family released on 11 December 2025, replacing GPT-5.1. It comes as three modes: instant for speed, thinking, and Pro. The last two reason through problems before replying, and a coding version called GPT-5.2-Codex was released too.
How do the modes differ?
Instant is built for fast, efficient answers. Thinking and Pro both reason first, and Pro spends more reasoning time and compute on each request. Pro also has two settings, standard and extended, which use different amounts of processing power and time.
Can I download the model?
No. The weights are proprietary. OpenAI offers it through ChatGPT and Microsoft Copilot, and documents it on its own product page.
Why did OpenAI release it when it did?
Reports said an internal "Code Red" memo, sparked by Google's Gemini 3 Pro, moved the launch forward from late December. OpenAI executives disputed that reading; Fidji Simo said months of work preceded it but that the extra push helped finish deployment.
What came after it?
GPT-5.3-Codex followed on 5 February 2026, and GPT-5.4 reached the public on 5 March 2026.
Good reading and sources
- Official websiteOfficial site
- GPT-5.2 is OpenAI’s latest move in the agentic AI battleThe Verge, Article
- OpenAI fires back at Google with GPT-5.2 after ‘code red’ memoTechCrunch, Article
- OpenAI Launches GPT-5.2 as It Navigates ‘Code Red’Wired, Article
The facts on this page come from the Wikipedia article "GPT-5.2", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryLLM rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.