Anthropic Launches the Claude 4 Model
On May 21, 2025, Anthropic released the much-anticipated Claude 4 model, hailed as the “world’s best coding model.” This release included two variants, Claude 4 Opus and Claude 4 Sonnet, which both excelled on industry-standard benchmarks. According to Anthropic’s data, Claude Opus achieved a score of 72.5% on the SWE-bench, while Claude Sonnet slightly outperformed with 72.7%. Additionally, Claude Opus marked 43.2% on the Terminal-bench.
The Claude 4 models introduced innovative features such as a hybrid operation mode, offering nearly instantaneous response times along with extended thinking capabilities. Notably, these models are capable of autonomously engaging in complex tasks for up to seven hours. They also feature a 200K token context window and improved memory functions. The models are designed to utilize a range of tools, such as web searching and code execution, even during extended operations, allowing for robust parallel tool use. Safety remains a priority, with Opus 4 earning a “Level 3” classification on Anthropic’s four-point safety scale.
Dario Amodei, Anthropic’s CEO, described this release as their “most intelligent model yet.” Apart from standalone availability, the models are now integrated with GitHub Copilot, broadening their utility in software development environments. Pricing for the Opus model is set at $15 for basic usage and $75 per million tokens for expansive needs, while Sonnet offers a more budget-friendly option at $3 and $15 respectively.
Today marked a relatively significant day in the AI field, primarily dominated by the launch of Claude 4, with no other major announcements reported.