Anthropic announced Claude 2 on July 11, 2023, as an update to its flagship model with higher benchmark scores on bar exam (76.5% vs 73%), coding (71.2% vs 56%), and math tasks (88% vs 85.2%), but no change to the 100k-token context window. The company characterized Claude 2 as a "tweaked version" of Claude 1.3 rather than an architectural redesign, trained on more recent data including early 2023 web content and third-party licensed datasets. The model is available in limited beta via API and web interface at unchanged pricing (~$0.0465 per 1,000 words), with early pilots from Jasper and Sourcegraph. For financial institutions, this signals continued iterative development in the frontier model space without step-function capability leaps, marginal cost structures holding steady, and embedded safety and liability concerns (data regurgitation, jailbreaks, hallucination) that Anthropic acknowledges but has not resolved.
272 other tracked developments touching the same claims.