Trusted by users from 10,000+ companies
Claude Sonnet 4.6 competes with top models, often beating them, in important benchmarks. Performance-to-cost ratio is among some of the best AI models.
Achieving a massive 79.6% on SWE-bench Verified (and up to 80.2% with optimized prompting), Sonnet 4.6 demonstrates elite software engineering capabilities, solving real-world GitHub issues with deeper reasoning.
With a score of 89.9% on GPQA Diamond, the model exhibits graduate-level expertise in the sciences, proving its ability to answer extremely challenging questions where even human experts often struggle.
In the domain of computer use, Sonnet 4.6 scores 72.5% on OSWorld-Verified, effectively matching the state-of-the-art Claude Opus 4.6 and showcasing its ability to navigate and operate computer interfaces just like a human user.
Streamline your corporate workflows with "Real-World Finance" capabilities that can research data and generate polished spreadsheets, slide decks, and Word documents. This model can synthesize raw data into structural deliverables.
Claude Sonnet 4.6 is packed with advanced features designed to handle the most demanding technical and creative challenges.
Allows the model to spend more compute time reasoning through complex tasks before responding, providing performance boost.
Enables the model to interact directly with computer GUIs (mouse and keyboard) to navigate websites and applications autonomously.
Delivers strong performance across high and mid-resource languages, due to MMMLU benchmark excellence.
Capable of solving complex software engineering tasks with a 79.6% success rate on real-world repositories.
Outperforms previous models in understanding scientific charts and figures, scoring 77.4% on CharXiv Reasoning with cropping tools.
Handles massive amounts of information with ease, maintaining 90.3% retrieval accuracy even in 256k token contexts.
Solves prestigious high school mathematics competition problems with a 95.6% accuracy on AIME 2025.
Specializes in researching SEC filings and financial data, achieving state-of-the-art accuracy in information retrieval for analysts.
Released under ASL-3 standards, it features industry-leading resistance to jailbreaks and prompt injections while minimizing refusals.
