Summary
Highlights
GPT 5.1 is introduced as a faster, more accurate, and more conversational model with an improved personality. The update includes two main versions: 5.1 Instant for quick, conversational answers and 5.1 Thinking for more complex problems, which now manages its thinking time more efficiently. The updated model addresses feedback regarding GPT-5's perceived 'boring' personality, aiming for a warmer and more intelligent interaction.
OpenAI has focused on making it easier to shape ChatGPT's personality in GPT 5.1. The video demonstrates how 5.1 provides more conversational and personal responses compared to its predecessor. It also showcases a significant improvement in instruction following, where GPT 5.1 adheres strictly to user commands, such as responding with a specific word count, which GPT-5 struggled with.
For the first time, GPT 5.1 Instant incorporates adaptive reasoning, allowing it to decide when to 'think' before responding to challenging questions. This results in more thorough and accurate answers while maintaining speed. Benchmarks show that GPT 5.1 is better at calibrating its thinking time based on question difficulty, leading to faster responses for easy questions and more extensive processing for complex ones, with significant improvements in math and encoding evaluations.
Box's benchmarks reveal massive improvements in enterprise use cases with GPT 5.1, particularly in reducing latency for document tasks and enhancing accuracy in document extraction, tabular data, and handwriting recognition. GPT 5.1 now demonstrates lower time to first token (TTFT) and higher accuracy across various complex document queries. The rollout of GPT 5.1 Instant and Thinking begins immediately for paid users (Pro, Plus, Go, Business), followed by free and logged-out users, with early access for Enterprise and .edu plans.
GPT 5.1 refines the ability to customize ChatGPT's style and tone, adding new personalities like 'professional,' 'candid,' and 'quirky' alongside existing options. The video also verifies that despite setting memory preferences, the model still occasionally uses M dashes, which is a minor frustration. Additionally, GPT 5.1 is significantly better at coding, especially front-end development, demonstrating increased accuracy in benchmarks like Swebench verified with extended thinking tokens.
GPT 5.1 is also released on the API platform, offering developers access to speed, accuracy, and token efficiency improvements. New features include extended prompt caching for up to 24 hours and the option to disable 'reasoning effort' for simpler tasks. The enhanced coding capabilities, particularly for front-end development, provide developers with a more powerful tool for building applications.