Claude 4 Opus recognized as the world’s best for coding, surpassing Gemini and OpenAI o3

Anthropic presented the 4th generation of Claude artificial intelligence. I am pleasantly shocked! The new lineup includes 2 models – Opus 4 and Sonnet 4. Both models demonstrate a fundamentally new level of work with programming and agent tasks.

Claude Opus 4 is positioned as the world’s best model for working with code. The main feature is the ability to maintain stable performance when executing complex long-term tasks. The model can continuously work on one task for hours, maintaining focus on the original context.

Claude Sonnet 4 represents a significant update to the previous version 3.7. The model shows improved capabilities in programming and logical thinking, while more accurately following user instructions.

Claude Sonnet can be tried for free, while Opus requires a subscription. Both models refuse to execute complex tasks 65% less often compared to their predecessors. Technologically, the models support extended thinking – they alternate analysis with the use of web search tools and execute tasks in parallel.

Opus 4 demonstrates deep understanding of complex codebases, while Sonnet 4 is optimized for everyday programming tasks. GitHub is already testing integration in their Copilot product. I have already tried both models and can say that the new models work more accurately and faster. Where previously there were several attempts, now everything is much faster and more accurate. I am really surprised!!!

Claude 4 Opus is right now the world’s best for coding, according to the SWE-bench benchmark and surpassed both Gemini 2.5 Pro and OpenAI o3.

At the presentation of the new models, Anthropic director Dario Amodei also gave an interesting forecast that already next year we will see the emergence of a company worth $1 billion with 1 employee and artificial intelligence.