Anthropic has launched its latest artificial intelligence models, Claude 4, during its inaugural developer conference held on Thursday. The San Francisco-based firm introduced two new models, Claude Opus 4 and Claude Sonnet 4, showcasing enhanced features, including Extended Thinking capabilities that leverage tool usage. Opus 4 is positioned as a state-of-the-art model for coding, writing, and tool interactions, with Claude Code now broadly accessible and beta extensions available for integration in VS Code and JetBrains, along with its presence on GitHub.
Anthropic Unveils Claude 4 AI Models
In an official news release, Anthropic elaborated on the functionalities of the new models and the enhancements rolled out across its chatbot platform and application programming interface (API). The company emphasizes coding proficiency and agent-like features within its latest large language models (LLMs).
Both Opus 4 and Sonnet 4 are designed as hybrid models, offering two modes: one for near-instantaneous responses and another for Extended Thinking, allowing for more profound reasoning. Opus 4, touted as the company’s flagship model, was described as “the best coding model in the world,” having achieved a score of 72.5 percent on the SWE-Bench benchmark and 43.2 percent on the Terminal-Bench, both of which assess model coding capabilities.
Performance metrics for Claude 4 models on SWE-Bench
Photo Credit: Anthropic
Similarly, Claude Sonnet 4 has demonstrated significant improvements over its predecessor, achieving a score of 72.7 percent on SWE-Bench, categorized as SOTA. While it does not surpass Opus 4 in all areas, Anthropic asserts that Sonnet 4 offers a better balance of performance and efficiency.
In addition to performance enhancements, Claude Opus 4 features improved memory capabilities that enable it to maintain long-term task awareness. The company has also addressed prior issues related to models finding shortcuts to complete tasks. Both models are capable of tool usage during Extended Thinking, allowing them to oscillate between intrinsic reasoning and the exploration of external information, such as conducting web searches to refine their responses. Additional enhancements include the ability to use tools simultaneously and improved adherence to prompts.
The Opus 4 and Sonnet 4 models, equipped with both operational modes, are currently accessible to subscribers of Claude Pro, Max, Team, and Enterprise tiers. Meanwhile, Sonnet 4 is also available to free users. Developers have the option to access these LLMs via the Anthropic API, as well as on platforms such as Amazon Bedrock and Google Cloud’s Vertex AI. Pricing remains consistent with previous versions.
For developers, Opus 4 is priced at $15 (approximately Rs. 1,290) per million input tokens and $75 (around Rs. 6,440) per million output tokens. Conversely, Sonnet 4 is available for $3 (approximately Rs. 260) per million input tokens and $15 (approximately Rs. 1,290) per million output tokens.
In addition to the new models, Anthropic announced several new features, including the general availability of Claude Code. Initially introduced in February as a research preview, this agentic coding tool can perform a broad range of programming tasks. Beta extensions for this tool are now available in VS Code and JetBrains. Furthermore, the company plans to release a Claude Code software development kit (SDK) currently in beta on GitHub.
