Anthropic's Claude 3.7 Sonnet can write code, fix bugs, and play Pokémon
Anthropic has introduced Claude 3.7 Sonnet, an updated chatbot with improved programming and data analysis skills.
Here's What We Know
The new version of AI is able to solve complex tasks thanks to an increased context window of 128K tokens and extended thinking time before answering. This allows Claude 3.7 Sonnet to process information more accurately and successfully cope with PhD-level tests such as GPQA.
Although the model does not always take the first place in the ratings, its performance has increased significantly. For example, Claude 3.7 Sonnet has advanced much further in Pokémon Red compared to previous versions. Besides, AI has become more useful for programmers: it can find errors in code, create software solutions, and even interact with GitHub to help edit and test complex projects.
However, increased intelligence also carries risks. Internal tests showed that Claude 3.7 Sonnet violated Anthropic's security policies three times more often than version 3.5, although the overall rate remained low at 0.6%. The AI was also able to carry out a cyberattack in the test environment, rewriting the code and stealing data. The public version has strict restrictions to prevent such scenarios.
Claude 3.7 Sonnet is available free of charge with basic features, while advanced capabilities, including deep analysis, require a subscription.
We would like to remind you that Perplexity has added Deep Research to its free chatbot.
Source: Anthropic
