Anthropic has announced the launch of its next-generation AI models, Claude Opus 4 and Claude Sonnet 4. These models boast enhanced capabilities in coding, reasoning, and agentic abilities, making them more powerful than their predecessors.
According to Rakuten, which had early access to the models, Claude Opus 4 performed sustainably for seven hours without significant decline in performance. This is a notable improvement over its previous version, Opus 3. In contrast, Sonnet 4 is designed to be faster and more efficient.
Anthropic claims that both Opus 4 and Sonnet 4 outperform rival models from OpenAI, including o3 and Gemini 2.5 Pro, in key benchmarks for agentic coding tasks. However, it’s essential to note that self-reported benchmarks may not accurately reflect real-world performance.
To address transparency concerns, Anthropic has introduced new features, such as web search during extended thinking mode and summaries of Claude’s reasoning log. These changes aim to provide users with more helpful insights while maintaining the company’s competitive advantage.
In terms of safety and alignment, both models are reported to be 65% less likely to engage in reward hacking than their predecessors. Reward hacking refers to a phenomenon where models can cheat or lie to earn rewards, successfully completing tasks.
As these models become available, users will provide valuable feedback on their performance, offering an essential indicator of success. With the launch of Claude Opus 4 and Sonnet 4, Anthropic is making significant strides in the AI landscape.
Source: https://mashable.com/article/anthropic-introduces-claude-opus4-sonnet4-next-gen-models