OpenAI Unveils GPT-5.1 for Developers

OpenAI has launched GPT-5.1 through its API platform. This model features Adaptive reasoning, which dynamically adjusts reasoning time according to task complexity, maximizing speed and token efficiency on simple tasks while maintaining high reliability on complex tasks.

In particular, it introduces a 'no reasoning' mode (reasoning_effort: 'none') that delivers high intelligence and excellent tool-calling performance even for latency-sensitive tasks. Developers can directly control the balance between speed, cost, and intelligence by setting the reasoning intensity to 'none', 'low', 'medium', or 'high' depending on the nature of the task.

To improve coding performance, OpenAI collaborated with major startups such as Cursor, Cognition, and Augment Code to enhance the model's controllability and code quality. It also introduced the apply_patch tool for more reliably editing code, and a new shell tool that allows the model to directly execute commands.

Key performance and efficiency updates are as follows:

• Extended prompt caching: Cache retention has been extended to up to 24 hours, speeding up responses to follow-up questions and reducing costs.
• Overwhelming speed and efficiency: In tests by Balyasny Asset Management, it achieved 2-3x faster speed and about half the token usage compared to GPT-5.
• Reduced latency: For simple npm command requests, tasks that took the existing GPT-5 10 seconds are now completed in about 2 seconds.