Anthropic Launches Claude 2 With Improved Coding and Reasoning Skills
Anthropic releases Claude 2, an upgraded AI chatbot that scores higher on math, coding, and professional exams than its predecessor. The new model is currently available in beta across the U.S. and U.K.
Anthropic launches Claude 2, the second generation of its commercial AI chatbot, making the upgraded model available in beta in the U.S. and U.K. The system builds upon the foundation of Claude 1.3 and handles the same core tasks like document search, summarization, writing, and coding. Several businesses, including Jasper and Sourcegraph, already pilot the new model through a paid API that maintains the previous pricing structure.
The new chatbot shows significant improvements in reasoning, math, and programming capabilities. Claude 2 achieves a 76.5% score on the bar exam multiple choice section and passes the U.S. Medical Licensing Exam. It also demonstrates stronger coding skills by scoring 71.2% on the Codex Human Level Python coding test and correctly answers 88% of grade-school-level math problems.
Anthropic trains Claude 2 on more recent data from early 2023, incorporating websites, licensed third-party datasets, and user-supplied information that includes non-English text. The company describes the release as a continuous, iterative improvement rather than a complete architectural overhaul. Unlike some competitors, Claude 2 does not search the live web, but Anthropic prioritizes real-world deployment to monitor usage, improve performance, and understand system limitations.