OpenAI releases GPT-5.5: comprehensive upgrade of programming, knowledge work and scientific research capabilities
OpenAI released GPT-5.5, achieving significant improvements in the three major areas of programming (Terminal-Bench +25.4%), knowledge work, and scientific research, while maintaining latency consistent with the previous generation.
OpenAI released GPT-5.5 on April 23, which is the latest upgraded version of the GPT-5 series. GPT-5.5 achieves significant capability improvements in the three core areas of programming, knowledge work, and scientific research, while maintaining latency performance that matches GPT-5.4.
In terms of programming capabilities, GPT-5.5's score on Terminal-Bench increased by 25.4%, and it also made significant progress on SWE-Bench. The new version also optimizes the quality and consistency of code generation, reducing the number of iterations on complex programming tasks.
In terms of knowledge work, GPT-5.5 has improvements in tasks such as document analysis, report generation, and information synthesis. The model can better understand long-form context and perform more stably in cross-document reasoning and multi-step analysis tasks.
The improvement in the field of scientific research deserves particular attention. GPT-5.5's scores on multiple scientific benchmarks demonstrate its potential in assisting research, including scenarios such as literature review, experimental design, and data analysis.
API pricing remains unchanged: GPT-5.5 continues the pricing system of GPT-5.4, and developers can obtain stronger capabilities at the same cost. In addition, OpenAI also launched the GPT-5.5 Pro version, designed for enterprise-level scenarios that require higher inference depth and longer output length.
The release of GPT-5.5 continues OpenAI’s rhythm of model iteration—capacity improvements every 3-4 months, while the price remains unchanged. This is good news for the developer ecosystem, but it also means that the competition in the field of large models is still accelerating. For Chinese developers, the differentiated positioning of GPT-5.5 and DeepSeek V4 series is clearer: the former focuses on the in-depth capabilities of Agent and programming scenarios, while the latter focuses on ultimate cost-effectiveness and open source flexibility.
Reviews