OpenAI has launched GPT-5.2, the latest evolution in the GPT-5 lineup, with enhancements in reasoning, tool use, long-context management, and spreadsheet/presentation creation. Accompanying the release are updated documentation, an improved system card, and guidance on integrating the new models in ChatGPT and the API.

GPT-5.2 is OpenAI’s newest top-tier model in the GPT-5 series, refining reasoning, tool use, long-context handling, and safety. System card data reveals enhanced robustness against jailbreaks and prompt injections with reduced hallucination rates, while documentation emphasizes improved spreadsheets, slides, and coding—alongside new controls for task compaction and reasoning effort.

Why GPT-5.2 is significant now

GPT-5.2 aims to unlock more everyday economic value: envision faster spreadsheet generation, clearer presentations, improved front-end code creation, enhanced image comprehension, and more dependable multi-step projects. For many teams, it's not just about benchmarks—but fewer restarts, more succinct reasoning, and improved tool integration right from the start.

What’s new at a glance

• Model tiers. GPT-5.2 is available in Instant (quick), Thinking (deeper reasoning), and Pro (higher compute for the most accurate responses). In ChatGPT, Auto can switch between Instant and Thinking for you.

• Capability enhancements. Upgrades include general intelligence, following instructions, token efficiency, multimodal vision, code/UI creation, tool integration, context management and spreadsheet comprehension.

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

• Model card data. GPT-5.2 Thinking shows improved outcomes on internal knowledge-work and science/math benchmarks compared to GPT-5.1.

Safety & reliability: what’s improved

OpenAI’s system card update indicates GPT-5.2 employs largely the same comprehensive safety strategy as GPT-5/5.1, with fresh evaluations showing advancements in multiple areas:

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

• Jailbreak resistance: GPT-5.2 Thinking surpasses 5.1 Thinking on a StrongReject-style jailbreak assessment (higher equals safer). OpenAI

• Prompt injection: Significant improvements on Agent JSK and PlugInject robustness, essentially mastering those evaluations. OpenAI

• Vision safety: On joint text+image examinations (hate, extremism, illicit content, attack planning, self-harm), 5.2 versions perform as well as or better than previous models. OpenAI

• Hallucinations: With browsing activated, GPT-5.2 Thinking reduces average hallucination rates compared to 5.1; across fields such as business/marketing, finance/tax, legal/regulatory, academic writing, current events, it maintains below a 1% false claim rate. OpenAI+1

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

Training & model specifications (what teams need to know)

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

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

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

Practical adoption

In ChatGPT (teams & enterprise)

• Choose GPT-5.2 Auto to let the system select between Instant or Thinking. You can manually pick Instant for fast tasks or Thinking for intricate work; you’ll see a streamlined reasoning display during extensive runs. OpenAI Help Center

In the API

• Utilize 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 for more challenging tasks), activate tools (File/Image/Code), and enable compaction for extended workflows. OpenAI Platform

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

Where 5.2 excels (specific tasks)

• Create spreadsheet models and financial summaries with fewer manual corrections.

• Produce and iterate on slide decks with a coherent framework and speaker notes.

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

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

FAQs

What’s the primary feature of GPT-5.2?
An expansive upgrade in reasoning and reliability: enhanced tool use, code/UI creation, spreadsheet/presentation improvements, and reduced hallucination rates—especially with browsing enabled. OpenAI Platform

How does GPT-5.2 differ from GPT-5.1?
It offers improved instruction adherence, accuracy, token efficiency, multimodal vision, tool integration, and introduces compaction and higher reasoning effort options for lengthy tasks. OpenAI Platform

What about safety?
OpenAI states the safety strategy aligns with GPT-5/5.1, yielding better outcomes on jailbreak, prompt-injection, vision-safety and hallucination evaluations in the system card update. OpenAI

Why is diverse training important?
Expansive, filtered datasets along with RL-based reasoning assist models to generalize and prevent policy circumvention, while classifiers mitigate risks of sensitive content. OpenAI

Where can developers learn to transition?
Consult the Using GPT-5.2 guide and model reference for endpoints, features, pricing, and version stability. OpenAI Platform