OpenAI launches ChatGPT 5.2: Key improvements in new version

The update is available in three versions – Instant, Thinking, and Pro – designed for different levels of reasoning depth.

Clear and consistent reasoning

OpenAI showcases tests where GPT-5.2 reaches or surpasses professional-level performance across dozens of categories. On the GDPval benchmark, which covers 44 professional domains (from finance to consulting), the Thinking version outperforms specialists in nearly 75% of tasks, including creating presentations, analyzing data, and preparing documents.

The main improvement is the model’s ability for multi-step reasoning. GPT-5.2 Thinking can:

• Break complex questions into parts;

• Plan a path to a solution;

• Maintain a logical chain;

• Produce results resembling a complete professional project.

In other words, the model no longer just "predicts phrases" but builds a coherent reasoning trajectory. This is reflected in clearer explanations, stability over long dialogues, and logic that doesn’t break down over time.

Expanded memory: working with hundreds of thousands of words

Improved memory directly enhances reasoning quality. OpenAI states that GPT-5.2 Thinking almost perfectly tracks information across vast text datasets.

Users can now load into ChatGPT:

• Project folders;

• Large reports;

• Legal documents;

• Scientific articles;

• Even with a full set of unfinished drafts, the model maintains the full context.

This significantly helps:

• Students – for comparative analysis of multiple texts;

• Business owners – for examining finances and strategies;

• Professionals – for handling complex multi-file projects.

Memory improvements also extend to visual tasks: GPT-5.2 better analyzes diagrams, charts, interfaces, and complex or blurry images.

Reliability: fewer mistakes, higher accuracy

The third key improvement is stability. According to OpenAI:

• Hallucinations have decreased by 30% compared to GPT-5.1.

• The model is more trustworthy in tasks requiring accuracy.

Improvements are also noticeable in ChatGPT’s built-in tools. For example, in one test, GPT-5.2 correctly handled a complex scenario involving a delayed flight, transfers, overnight stay, and a medical chair request – whereas GPT-5.1 failed mid-process.

OpenAI also focused on safer interactions. GPT-5.2 responds more appropriately to topics related to emotional states or mental health and provides more balanced advice.

Availability

The model will gradually roll out in ChatGPT Plus, Pro, Go, Business, and Enterprise subscriptions.

Access to GPT-5.1 will be closed in three months.

Experts believe this update could be a turning point – not flashy, but a calm, mature improvement for users who want AI to think as carefully as it communicates.