We are excited to introduce GPT-5.2, our most advanced modern series for professional knowledge work.
On average, ChatGPT Enterprise users report saving 40-60 minutes each day, while heavy users save over 10 hours per week. We built GPT-5.2 to help people get even more value from their work. It’s now better at:
- making spreadsheets
- building presentations
- writing code
- analyzing images
- understanding longer documents
- using tools
- coordinating complex projects
GPT-5.2 sets new standards on many benchmarks, including GDPval, where it performs better than industry professionals on well-defined tasks across 44 different jobs.
Companies like Notion, Box, Shopify, Harvey, and Zoom have found that GPT-5.2 excels at long-term reasoning and tool use. Databricks, Hex, and Triple Veil found it especially strong in data science and document analysis. Charley Labs, JetBrains, and Augment Code report that GPT-5.2 offers top performance in coding, with clear improvements in interactive coding, code reviews, and bug finding.
In ChatGPT, GPT-5.2 Instant Thinking and Pro will start rolling out today for paid plans. For developers, these models are already available in the API.
Overall, GPT-5.2 offers major improvements in general intelligence, understanding long documents, and working with images. This makes it better at handling complex real-world tasks from start to finish than any earlier model.
Model Effectiveness
Economically Valuable Tasks
GPT-5.2 thinking is our best model so far for actual professional tasks on GDPval, which measures performance on specific knowledge work across 44 jobs. GPT-5.2 thinking achieved a new top score and is our first model to match or exceed human experts. It outperformed or matched top professionals in 70.9% of cases, according to expert judges. These tasks include creating presentations, spreadsheets, and similar work. GPT-5.2 completed these tasks over 11 times faster and at less than 1% of the cost of expert professionals. These estimates are based on past data, and actual speed in ChatGPT may vary.
A GDP well judge praised one remarkable output, saying, “It is an exciting and noticeable leap in output quality. It appears to have been done by a professional company with staff. It has a surprisingly well-designed layout and advice for both deliverables, though with one, we still have some minor array errors to correct.”
In our internal tests for Junior Investment Banking Analyst tasks like building a 3-statement model for a Fortune 500 company or developing a leveraged buyout model, GPT-5.2 thinking scored 9.3% higher per task than GPT-5.1. The average score went up from 59.1 to 68.4.
Spreadsheets and slides made by GPT-5.2 thinking are more advanced and better formatted than before.
To use the new Spreadsheet and Presentation features in ChatGPT, you need a Plus, Pro, Business, or Enterprise plan and must select either GPT-5.2 Thinking or Pro. Some complex results may take a few minutes to generate.
Coding
GPT-5.2 rethinking reached a new high score of 55.6 on SWE Bench Pro, a tough test for Real Software Engineering. While the SWE Bench verifies only Python, SWE Bench Pro supports four languages and is designed to be harder, more resistant to contamination, and more relevant to industry needs.
On the SWE bench verified (not shown in the chart), GPT-5.2 thinking achieved a new high score of 80%.
In daily work, this means the model can more reliably:
- debug production code
- handle future requests
- Refactor large code bases
- deliver fixes from start to finish with less manual work
GPT-5.2 thinking also outperforms GPT-5.1 thinking in front-end software engineering. Early testers found it much better at front-end development and at operating complex or unusual UI tasks, especially those with 3D elements. This makes it a strong daily partner for engineers working across the stack.
GPT-5.2 instant update (February 10, 2026)
We are updating GPT-5.2 instant in ChatGPT and the API to improve reply style and quality.
Users should notice that responses are more balanced and better fit the conversation. The model also gives clearer, more relevant answers to advice-seeking and how-to questions and puts the most important information first more often.
Update on Thinking Time Settings for GPT-5.2 Thinking in ChatGPT (February 4, 2026)
Jan 10, 2026: We lowered the standard and light thinking time after seeing that users prefer faster responses. During this update, the extended thinking setting for GPT-5.2 was unintentionally set to a lower value, but we have now fixed it.
February 3, 2026: We made another small reduction to standard thinking time based on testing.
Feb 4, 2026: We are restoring the extended thinking level for GPT-5.2 thinking to its previous setting, fixing the accidental reduction from January. Extended time is now back to its earlier level.
We regularly adjust the default thinking time of our reasoning models. These changes result from ongoing tests to find the best balance between answer quality and user response speed.
The thinking level toggle introduced in September 2025 gives users more options beyond standard. It lets them pick the right-thinking level for their question, whether they want quicker, lighter responses or more detailed reasoning when depth and accuracy are important.
Thinking time is not directly comparable among different models. Each model is tuned separately to work best for users. We will keep changing these settings as models change and will continue to give users clear controls when there are important trade-offs to consider.
