OpenAI has released GPT-5.2, the next step in the GPT-5 family, with upgrades in reasoning, tool use, long-context handling, and spreadsheet/presentation creation. The launch comes with fresh docs, a system-card update, and guidance on how to adopt the new models in ChatGPT and the API.

GPT-5.2 is OpenAI’s latest flagship in the GPT-5 series, improving reasoning, tool use, long-context handling and safety. System-card data shows stronger jailbreak and prompt-injection robustness and lower hallucination rates, while docs highlight better spreadsheets, slides and coding—plus new compaction and reasoning-effort controls for longer tasks.

Why GPT-5.2 matters now

GPT-5.2 aims to unlock more day-to-day economic value: think faster spreadsheet builds, clearer slides, stronger front-end code generation, better image understanding, and more reliable multi-step projects. For many teams, the headline isn’t just benchmarks—it’s fewer restarts, tidier reasoning, and improved tool calling out of the box.

What’s new at a glance

• Model tiers. GPT-5.2 ships as Instant (fast), Thinking (deeper reasoning), and Pro (more compute for consistently best answers). In ChatGPT, Auto can switch between Instant and Thinking for you.

• Capability gains. Improvements span general intelligence, instruction following, token efficiency, multimodal vision, code/UI generation, tool calling, context management and spreadsheet understanding.

• Context + memory features. New compaction and reasoning-effort controls (including an xhigh setting) help manage long tasks without losing the thread.

• Model card data. GPT-5.2 Thinking shows higher results on internal knowledge-work and science/maths benchmarks versus GPT-5.1.

Safety & reliability: what’s improved

OpenAI’s system-card update says GPT-5.2 uses largely the same comprehensive safety approach as GPT-5/5.1, with new evaluations showing gains in multiple areas:

• Disallowed content: Production Benchmark scores remain strong; notable improvements on self-harm, mental health and emotional reliance evals. OpenAI

• Jailbreak resistance: GPT-5.2 Thinking outperforms 5.1 Thinking on a StrongReject-style jailbreak eval (higher = safer). OpenAI

• Prompt injection: Big jumps on Agent JSK and PlugInject robustness, essentially saturating those evals. OpenAI

• Vision safety: On combined text+image tests (hate, extremism, illicit, attack planning, self-harm), 5.2 variants perform on par or better than predecessors. OpenAI

• Hallucinations: With browsing enabled, GPT-5.2 Thinking lowers average hallucination rates vs 5.1; across domains like business/marketing, financial/tax, legal/regulatory, academic essays, current events, it stays under 1% incorrect-claim rate. OpenAI+1

OpenAI also reiterates training and filtering practices: diverse datasets (public web, licensed/partnered sources, and user-provided data) with rigorous filtering to remove personal or sensitive content; reasoning models are trained to “think before answering”. OpenAI

Training & model specs (what teams need to know)

• Data sources & processing: public, partner, and user-contributed data with safety classifiers and filtering to mitigate risks (e.g., personal data, sexual content involving minors).

• Knowledge cut-off: 31 Aug 2025 for GPT-5.2 model docs (check live docs for updates).

• Context & outputs: up to 400k tokens context and 128k max output tokens on the flagship API model.

Practical adoption

In ChatGPT (teams & enterprise)

• Pick GPT-5.2 Auto to let the system route to Instant or Thinking. You can manually choose Instant for quick tasks or Thinking for complex work; you’ll see a slimmed-down reasoning view during deeper runs. OpenAI Help Center

In the API

• Use the Responses API (or Chat Completions) with gpt-5.2, gpt-5.2-pro, or gpt-5.2-chat-latest.

• Set reasoning effort (try high → xhigh on harder tasks), enable tools (File/Image/Code), and turn on compaction for long-running workflows. OpenAI Platform

• Pin a snapshot (e.g., gpt-5.2-2025-12-11) for consistent behaviour across sprints. OpenAI Platform

Where 5.2 shines (concrete jobs-to-be-done)

• Build spreadsheet models and financial summaries with fewer manual fixes.

• Generate and iterate slide decks with coherent structure and speaker notes.

• Front-end prototypes (UI scaffolds/components) with cleaner code.

• Long-context reviews (RFPs, contracts, research) with tool-calling for citations.

FAQs

What’s the main feature of GPT-5.2?
A broad upgrade in reasoning and reliability: better tool use, code/UI generation, spreadsheets/slides, and lower hallucination rates—especially when browsing is enabled. OpenAI Platform

How does GPT-5.2 differ from GPT-5.1?
It improves instruction following, accuracy, token efficiency, multimodal vision, tool calling, and adds compaction and higher reasoning effort options for long tasks. OpenAI Platform

What about safety?
OpenAI says the safety approach mirrors GPT-5/5.1, with better results on jailbreak, prompt-injection, vision-safety and hallucination tests in the system-card update. OpenAI

Why is diverse training important?
Broader, filtered datasets plus RL-based reasoning help models generalise and resist policy bypasses, while classifiers reduce sensitive content risks. OpenAI

Where can developers learn to migrate?
See the Using GPT-5.2 guide and model reference for endpoints, features, pricing and snapshots. OpenAI Platform