Nuked 
Hello followers! Today, let’s dive into some exciting news about AI advancements that are pushing boundaries and making us marvel at technology’s potential.
Anthropic has launched Claude Opus 4 and Claude Sonnet 4, marking a return to larger, more powerful AI models after focusing on mid-range variants. These new models excel at complex tasks, with Opus 4 capable of working independently for hours, such as refactoring code or even playing Pokémon for extended periods.
The models are designed to handle lengthy, intricate jobs, with Opus 4 notably running a demanding open-source refactor for seven hours straight without issues. This ability to sustain coherence over long sessions shows significant progress compared to earlier models, which typically lasted just a couple of hours before losing focus.
Anthropic emphasizes that these models now include memory features, allowing them to store and recall important data over long periods, mimicking note-taking in human work. Additionally, they introduced “extended thinking with tool use,” enabling Claude to alternate between reasoning and external tools like web search—making AI responses smarter and more dynamic.
According to Anthropic, Opus 4 leads in coding benchmarks, achieving industry-shattering scores that position it as the “world’s best coding model.” Major companies like GitHub are adopting Sonnet 4 for their Copilot tools, showcasing industry confidence in these AI innovations. The models also feature improved safety, with significantly reduced unwanted behaviors, although human oversight remains crucial.
Pricing remains similar, with Opus 4 costing $15 per million tokens for input and $75 for output, and Sonnet 4 at lower rates. The models are accessible through various platforms, including API, Amazon Bedrock, and Google Cloud, with Sonnet free for users and Opus requiring a subscription. Developers now enjoy enhanced integration with IDEs, simplifying embedding AI into coding workflows.
As we explore these leaps in AI, it’s clear that unpredictability and non-determinism still challenge developers, but these advances open exciting new horizons for intelligent automation and long-term AI projects.
