Anthropic Launches Claude 2 AI Model With Expanded Context And Better Reasoning
Anthropic releases Claude 2, an upgraded AI assistant that handles 100,000 tokens per prompt and achieves higher scores on coding and academic benchmarks. The new model is available through an API and a public beta website for users in the US and UK.
Anthropic officially unveils Claude 2, a new artificial intelligence model that delivers enhanced performance in coding, math, and reasoning tasks. The upgraded system scores 76.5% on the Bar exam multiple choice section and achieves a 71.2% score on a Python coding test, representing significant jumps from the previous version. Businesses access the Claude 2 API at the same price point as Claude 1.3, while individual users in the US and UK explore the capabilities through a new public-facing beta website.
The new model handles up to 100,000 tokens per prompt, enabling users to upload hundreds of pages of technical documentation or entire books for analysis. Claude 2 also generates longer output documents in a single response, seamlessly producing memos, letters, or stories up to a few thousand tokens. These expanded context windows transform the AI into a highly capable research assistant that maintains context over massive amounts of text.
Safety remains a core focus for Anthropic, with Claude 2 proving to be twice as resistant to generating harmful outputs compared to its predecessor. Internal red-teaming evaluations combine automated testing with manual reviews to ensure the model avoids offensive or dangerous content. Anthropic plans to slowly and iteratively deploy further capability improvements to Claude 2 in the coming months.