GPT-5.2 is Here

What It Covers

OpenAI releases GPT-5.2, positioning it as a professional work model that scores 70.9% on economically valuable tasks, outperforming competitors on spreadsheets, presentations, and coding while reducing hallucinations by 30-40%.

Key Questions Answered

• •Professional Task Performance: GPT-5.2 achieves 70.9% on GDP Val benchmark measuring real business tasks like spreadsheet creation and presentations, up from 38.8% with GPT-5, with OpenAI emphasizing economic value over general capabilities throughout their messaging.
• •Coding Improvements: The model scores 55.6% on SWEBench Pro coding benchmark versus Opus 4.5's 52%, enabling more reliable debugging, feature implementation, and code refactoring with less manual intervention, though specialized coding models may still lead in some scenarios.
• •Long Context Handling: Performance on needle-in-haystack tests maintains above 90% accuracy at 256k context length compared to GPT-5.1's drop below 50%, enabling processing of massive enterprise codebases and documents without degradation.
• •Pro Version Differentiation: GPT-5.2 Pro demonstrates extended reasoning capabilities, spending significantly more time on complex problems and understanding implicit constraints beyond literal requests, though standard version suffers from slow speed that limits daily usage.