OpenAI has launched GPT-5.1, an advanced language model now available via API, maintaining GPT-5 pricing. This update introduces gpt-5.1-codex and gpt-5.1-codex-mini variants designed for long programming tasks. Enhanced features include prompt caching for 24 hours, improving speed and reducing costs for recurring queries. Benchmarks show that GPT-5.1 offers moderate improvements, scoring 76.3% on the SWE-bench, up from 72.8%. New functionalities include a “No Reasoning” mode for faster responses and an “apply_patch” tool for code manipulation, enhancing developer automation. The model displays improved warmth in responses but raises safety concerns, as it has become less effective at managing sensitive topics such as harassment and hate speech. Emotional dependency is another concern, with a notable decline in resistance. Additionally, the model’s ability to block jailbreak attempts has increased significantly. Overall, GPT-5.1 is a nuanced update aimed at improving user interaction and developer efficiency.
Source link
Share
Read more